Big Data Reduction Methods: A Survey
 15k Downloads
 16 Citations
Abstract
Research on big data analytics is entering in the new phase called fast data where multiple gigabytes of data arrive in the big data systems every second. Modern big data systems collect inherently complex data streams due to the volume, velocity, value, variety, variability, and veracity in the acquired data and consequently give rise to the 6Vs of big data. The reduced and relevant data streams are perceived to be more useful than collecting raw, redundant, inconsistent, and noisy data. Another perspective for big data reduction is that the million variables big datasets cause the curse of dimensionality which requires unbounded computational resources to uncover actionable knowledge patterns. This article presents a review of methods that are used for big data reduction. It also presents a detailed taxonomic discussion of big data reduction methods including the network theory, big data compression, dimension reduction, redundancy elimination, data mining, and machine learning methods. In addition, the open research issues pertinent to the big data reduction are also highlighted.
Keywords
Big data Data compression Data reduction Data complexity Dimensionality reduction1 Introduction
Big data is the aggregation of largescale, voluminous, and multiformat data streams originated from heterogeneous and autonomous data sources [1]. The volume is the primary characteristic of big data that is represented by the acquisition of storage spaces in largescale data centers and storage area networks. The massive size of the big data not only causes the data heterogeneity but also results in diverse dimensionalities in the datasets. Therefore, efforts are required to reduce the volume to effectively analyze big data [2]. In addition, big data streams are needed to be processed online to avoid lateral resource consumption for storage and processing. The second key characteristic of big data is velocity. The velocity refers to the frequency of data streams, which is needed to be abridged in order to handle big data effectively. For example, solar dynamics observatory generates excess of one terabytes data per day and the analysis of such a fast big data is possible only after reduction or summarization [3]. On the other hand, big data inherits the ‘curse of dimensionality.’ In other words, millions of dimensions (variables, features, attributes) are required to be effectively reduced to uncover the maximum knowledge patterns [4, 5]. For example, behavior profiles of the Internet users that mainly comprise of searches, pageviews, and clickstream data are sparse and high dimensional with millions of possible keywords and URLs [6]. Similarly, personal genomic highthroughput sequencing not only increases the volume and velocity of data but also adds to the high dimensionality of the data [7]. Therefore, it is imperative to reduce the high dimensions while retaining the most important and useful data.
Data reduction methods for big data vary from pure dimension reduction techniques to compressionbased data reduction methods and algorithms for preprocessing, clusterlevel data deduplication, redundancy elimination, and implementation of network (graph) theory concepts. Dimension reduction techniques are useful to handle the heterogeneity and massiveness of big data by reducing million variable data into manageable size [8, 9, 10, 11]. These techniques usually work at postdata collection phases. Similarly, cluster deduplication and redundancy elimination algorithms that remove duplicated data for efficient data processing and useful knowledge discovery are primarily postdata collection methods [12, 13, 14, 15]. Recently, the network theory concepts have also been employed for big data reduction [16, 17, 18]. The aforementioned methods first extract the semantics and linked structures from the unstructured datasets and then apply graph theory for network optimization. Conversely, some methods to reduce big data during the data collection process are also proposed in the recent literature [19, 20, 21]. In this study, we presented a detailed discussion of these data reduction methods.
This article presents a thorough literature review of methods for big data reduction. A few similar prior studies have also been conducted. However, these studies either present a generic discussion of big data reduction or discuss a specific group of relevant systems or methods. For example, the authors in [1] discussed the big data reduction to be the critical part of mining sparse, uncertain, and incomplete data. Similarly, the authors in [22, 23] argue big data reduction as the critical part of data analysis and data preprocessing. However, both of the studies lack in presenting discussion about specific systems and methods for big data reduction. The authors in [4] discussed big data reduction issue specifically by focusing on dimension reduction, whereas the authors in [24] emphasized on the data compression. However, a wide range of methods remain unexplored. Currently, there is no specific study in the literature that addresses the core issue of big data reduction. Therefore, we aim to present a detailed literature review that is specifically articulated to highlight the existing methods relevant to big data reduction. In addition, some open research issues are also presented to direct future researchers.

A thorough literature review and classification of big data reduction methods are presented.

Recently proposed schemes for big data reduction are analyzed and synthesized.

A detailed gap analysis for the articulation of limitations and future research challenges for data reduction in big data environments is presented.
The article is structured as follows: Sect. 2 discuses the complexity problem in big data and highlights the importance of big data reduction. The taxonomical discussion on big data reduction methods is presented in Sect. 3. The discussion on open issues and future research challenges is given in Sect. 4, and finally, the article is concluded in Sect. 5.
2 Big Data Complexity and the Need for Data Reduction

Volume The data size characterizes the volume of big data. However, there is no agreed upon definition of big data which specifies the amount of data to be considered as ‘big’ on order to meet the definition of big data. However, a common sense is developed in research community who consider any data size as big in terms of volume which is not easily processable by underlying computing systems. For example, a large distributed system such as computing clusters or cloudbased data centers may offer to process multiple terabytes of data but a standalone computer or resource constrained mobile devices may not offer the computational power to process even a few gigabytes of data. Therefore, the volume property of big data varies according to underlying computing systems.

Velocity The velocity of big data is determined by the frequency of data streams which are entering in big data systems. The velocity is handled by big data systems in two ways. First, the whole data streams are collected in centralized systems, and then, further data processing is performed. In the second approach, the data streams are processed immediately after data collection before storing in big data systems. The second approach is more practical; however, it requires a lot of programming efforts and computational resources in order to reduce and filter the data streams before entering in big data systems.

Variety Big data systems collect data stream from multiple data sources which produce data streams in multiple formats. This heterogeneity in data sources and data types impacts the variety propertyrelated characteristics. Therefore, big data systems must be able to process multiple types of data stream in order to effectively uncover hidden knowledge patterns.

Veracity The utility of big data systems increases when the data streams are collected from reliable and trustworthy sources. In addition, the data stream collection is performed with compromising the quality of data streams. The veracity property of big data relates to reliability and trustworthiness of big data systems.

Variability Since all data sources in big data systems do not generate the data streams with same speed and same quality. Therefore, variability property enables to handle the relevant issues. For example, the elastic resource provisioning as per the requirements of big data systems.

Value The value property of big data defines the utility, usability, and usefulness of big data systems. This property tends more toward the outcomes of data analytics and data processing processes and is directly proportional to other 5Vs in big data systems.
The welldesigned big data systems must able to deal with all 6Vs effectively by creating a balance between data processing objectives and the cost of data processing (i.e., computational, financial, programming efforts) in big data systems.
Moreover, the complexity in big data systems emerges in three forms: (1) data complexity, (2) computational complexity, and (3) system complexity [28]. The data complexity arises due to multiple formats and unstructured nature of big data, which elevate the issue of multiple dimensions and the complex interdimensional and intradimensional relationships. For example, the semantic relationship between different values of the same attribute, for example, noise level in the particular areas of the city, increases the interdimensional complexity. Likewise, the linked relationship among different attributes (for example, age, gender, and health records) raises the intradimensional complexity issue. In addition, the increasing level of data complexity in any big data system is directly proportional to the increase in computational complexity where only the sophisticated algorithms and methods can address the issue. Moreover, the systemlevel complexity is increased due to extensive computational requirements of big data systems to handle extremely large volume, complex (mostly unstructured and semistructured), and sparse nature of the data. The extensive literature review exhibits that the big data reduction methods and systems have potential to deal with the big data complexity at both algorithms and systems level. In addition to data complexity, the big data reduction problem is studied in various other perspectives to articulate the effects and the need of data reduction for big data analysis, management, commercialization, and personalization.
Big data analysis also known as big data mining is a tedious task involving extraneous efforts to reduce data in a manageable size to uncover maximum knowledge patterns. To make it beneficial for data analysis, a number of preprocessing techniques for summarization, sketching, anomaly detection, dimension reduction, noise removal, and outliers detection are applied to reduce, refine, and clean big data [29]. The New York Times, a leading US newspaper, reports that data scientists spend 50–80% of the time on cleaning the big datasets [30]. The terms used in the industry for the aforementioned process are ‘data munging,’ ‘data wrangling,’ or ‘data janitor work.’ Another issue with the largescale highdimensional data analysis is the overfitting of learning models that are generated from large numbers of attributes with a few examples. These learning models fit well within the training data, but their performance with testing data significantly degrades [31].
Data management is another important aspect to discuss the big data reduction problem. The effective big data management plays a pivotal role from data acquisition to analysis and visualization. Although data acquisition from multiple sources and aggregation of relevant datasets improve the efficiency of big data systems, it increases the innetwork processing and data movement at clusters and data center levels. Similarly, the indexing techniques discussed in [26] enhance the big data management; however, the techniques come across data processing overheads. Although the conversion of unstructured data to semistructured and structured formats is useful for effective query execution, the conversion in itself is a time and resourceconsuming activity. Moreover, big data is huge in volume that is distributed in different storage facilities. Therefore, the development of learning models and uncovering global knowledge from massively distributed big data is a tedious task. Efficient storage management of reduced and relevant data enhances both the local learning and global view of the whole big data [32, 33]. Currently, visual data mining technique of selecting subspace from the entire feature spaces and subsequently finding the relevant data patterns also require effective data management techniques. Therefore, the reduction in big data at the earliest enhances the data management and data quality and therefore improves the indexing, storage, analysis, and visualization operations of big data systems.
Recently, businesses particularly the enterprises are turning into big data systems. The collection of large data streams from Web users’ personal data streams (clickstreams, ambulation activities, geolocations, and health records) and integration of those data streams with personalized services is a key challenge [34]. The collection of irrelevant data streams increases the computational burden that directly affects the operational cost of enterprises. Therefore, the collection of finegrained, highly relevant, and reduced data streams from users is another challenge that requires serious attention while designing big data systems. Currently, user data collection by third parties without explicit consent and information about commercialization is raising the privacy issues. The participatory personal data where users collect and mine their own data and participate for further utilization and customization of services in ubiquitous environments can address the issue of finegrained data availability for enterprises. Keeping in view the big data complexity, the need for big data reduction, and analyzing big data reduction problem in different perspective, we present a thorough literature review of the methods for big data reduction.
3 Big Data Reduction Methods
This section presents the data reduction methods being applied in big data systems. The methods either optimize the storage or innetwork movement of data or reduce data redundancy and duplication. In addition, some of the methods only reduce the volume by compressing the original data and some of the methods reduce the velocity of data streams at the earliest before entering in big data storage systems. Alternatively, some of the methods extract topological structures of unstructured data and reduce the overall big data using network theory approaches that are discussed as follows.
3.1 Network Theory
Aside from the influencebased assessment, similarity graph and collapsing simplicial complexes in network structures are used for topological network reduction in big datasets. Similarity graph is used to model the relationship among different datasets on the basis of their similarity matrix [17]. Further optimization of similarity graph is performed by merging every vertex with the maximal similarity clique (MSC), where a similarity clique (SC) is the collection of nodes with negligible distance. The notion is to convert a SC into MSC by adding an adjacent vertex that violates the properties of the SC. The graph is reduced when every MSC in the similarity graph is merged into a single node and the number of nodes and their edges are reduced in turn. Although the MSC is an efficient scheme to reduce topological networks, finding all of the MSCs in the similarity graph is quite challenging and computationally inefficient.
Moreover, the simplicial complexes are the multidimensional algebraic representations of large datasets [18]. The simplicial complex approach is applied over persistent homology, which is a method to compute topological features of large spaces at different levels of spatial resolutions. The persistent homology algorithm, as an alternate of hierarchical clustering, takes nested sequences of simplicial complexes and returns the summary of topological features at all levels. The concept of strong collapse that is keeping relevance information about all of the nodes in the simplicial complex is used to reduce the big data. The strong collapse is also useful for parallel and distributed computations, but the increasing computational complexities of maintaining relevance information of all of the nodes in complex prove to be the bottleneck. The concept of selective collapse algorithm is introduced to reduce the computational complexity of processing persistent homology. The proposed framework keeps computation traces of collapsing the complex using persistence algorithm proposed in [42]. In addition, the strong collapses are represented by forest that facilitates in easy processing of persistence across all nodes of the complex. Although the selective collapse algorithm is useful for the big data reduction, the empirical evidences are still lacking in the literature.
3.2 Compression
Big data compression methods
References  Methods  Description  Strengths  Weaknesses 

Yang et al. [45]  Spatiotemporal  The proposed method performs online clustering of streaming data by correlating similarities in their time series to share workload in the clusters. In addition, it performs temporal compression on each network node to reduce overall data  Performance enhancement in terms of data quality, information fidelity loss, and big data reduction  At least one time processing of whole data is performed in cloud environment which increases the operational cost 
Ackermann and Angus [44]  gZip  gZip is a compression tool developed that is being used for big data reduction to improve the resource efficiency for the IPactivity dataset in social science  It provides a light weight and simple file format; therefore, it has low computational complexity  gZip compresses one file at a time; therefore, massively parallel programming models like mapreduce must be used for performance gain in terms of computation time 
Jalali and Asghari [47]  AST  The AST is a novel method that is used to compress digital signal by performing selective stretching and wrapping methods. The technique is primarily based on selfadaptive stretch where more samples are associated with sharp features and less samples are associated with redundant coarse features  AST performs data compression of the signal extracted on frequency domain. The method also performs inverse transformation of the constructed signal  The method specifically works with big data involving signal processing. The generalization to other domain is a bottleneck in this research 
Wang et al. [48]  Compressed sensing  Compressed sensing is a compressible and/or sparse signal that projects a highdimensional data in lowdimensional space using random measurement matrix  The proposed scheme performs data acquisition and compression in parallel for improved performance as compared with Nyquist sampling theorybased compression methods  The probability of poor data quality and information fidelity loss increases when the analyses are performed on reduced and compressed data 
Brinkmann et al. [50]  RED encoding  RED encoding used to manage massively generated voluminous electrophysiology data  RED performs best when encoding invariant signals, and it provides high compression rate with improved computational speed in lossless compression of time series signals  The performance of the RED encoding methods degrades with high variance in signals 
Bi et al. [55]  Parallel compression  Parallel compression methods uses proper orthogonal decomposition method to compress data in order to effectively trace and extract useful features from the data  Balances between feature retention error and compression ratio Performs fast decompression for interactive data visualization  Due to noise, the standard deviation of error remains high in the dataset 
Monreale et al. [46]  Sketching  Sketching uses countmin sketch algorithm to compress vehicular movement data and achieve compact communication  Guarantees data reduction and preserves some important characteristics of the original data  The probability of information fidelity loss is more when sketching applied with inconsistent and noisy data stream 
Big data reduction in cloud environments is quite challenging due to multiple levels of virtualization and heterogeneity in the underlying cloud infrastructure. A spatiotemporal technique for data compression on big graph data in the cloud generates reduced datasets [45]. The technique performs online clustering of streaming data by correlating similarities in their time series to share workload in the clusters. In addition, it performs temporal compression on each network node to reduce the overall data. The proposed technique effectively meets the data processing quality and acceptable fidelity loss of the most of the application requirements. On the other hand, wireless sensor networks (WSNs) are generating large data streams at massive scales. The spatiotemporal data compression algorithm proposed in [46] ensures efficient communication, transmission, and storage of data in WSNsbased big data environment. The proposed approach not only reduces the size of transmitted data but also ensures prolonged network lifetime. The algorithm measures the correlation degree of sensed data, which determines the content of the data to be transmitted.
The authors in [44] proposed an efficient big data reduction scheme for the IPactivity dataset in social science. The techniques are based on compression method to utilize the standalone computer machines instead of largescale distributed systems, such as Hadoop and big table. The authors used eight core processors from a 32 core AMD Opteron Processor 5356 machine of 2.3 GHZ speed and 20 GB RAM. The 9 TB of loose text files was converted into a binary format for readability in social science settings. The methodology is based on three steps for: (1) information representation, (2) parallel processing, and (3) compression and storage. Big data reduction is performed in second and third steps. First, each of the data files is processed and converted into a corresponding HDFS file; then, mapreduce approach is used to link each individual HDFS file with the corresponding geographical locations and aggregate in lateral stages. In mapphase, the files are linked, whereas in the reduce stage, all of the resultant HDFS files are converted into a single HDFS file. Finally, the gzip algorithm was used with the compression level of five over the optimal data chunk size of 100,000 × 1. The results exhibited a significant amount of data reduction from 9.08 TB (raw data) to 3.07 TB (HDFS converted data) and 0.50 TB (compressed data). Although the proposed approach contributed significantly, it is only useful for the given dataset. Therefore, online data analysis for big data reduction remains challenging in streaming big data environments.
Compressive sensing theory is useful because lowdimensional space fully represents highdimensional data. The results showed that the proposed algorithm produced satisfactory results with and without compressed sensing. The compression was applied by the ratio of 1:3 and 1:5, and the experiments for human detection were performed from behind the brick and gypsum walls.
The authors in [50] proposed a novel file format called multiscale electrophysiology format (MEF) to manage massively generated electrophysiology data. The blockwise lossy compression algorithm (RED encoding) is used in addition to the cyclic redundancy check (CRC), encryption, and block index structure of the MEF. The RED encoding ensures high lossless compression rate and higher computational speed and is also able to adopt with statistical variation in the raw data, which is very important for nonstationary ECG signals. The experimental results showed that for 32 KHZ recordings with each block of 32,556 samples acquired in one second, the MEF obtained reasonably better compression rate. The authors recommend recording at least 2000 samples in each block for the maximum performance gain.
The amount of data generated in vehicular ad hoc networks is massive due to onboard monitoring of vehicles and relevant spatiotemporal data acquisition. The authors in [46] utilized sketching algorithm called countmin sketch algorithm to compress vehicular movement data and achieved compact communication. The algorithm maps frequency counters to compressed vectors using hash tables. Although the main focus of the proposed study is on privacy preservation, data reduction is also performed significantly.
The largescale scientific simulations generate a huge amount of data that widens the gap between the I/O capacity and computation abilities of highend computing (HEC) machines [51]. This bottleneck for data analysis raises the need for in situ analytics where simulation data are processed prior to the I/O. Although feasible, the in situ analysis of petascale data incurs computation overhead in the HEC machines. The authors proposed adaptive compression service for in situ analytics middleware to effectively utilize available bandwidth and to optimize the performance of the HEC during endtoend data transfer. Experimental results with gyrokinetic simulation (GKW) on 80node 1280 core cluster machine show that the compression ratio and available computational power are two main factors to achieve the maximum compression. The authors in [43] further proposed a framework called FlexAnalytics and profiled three compression algorithms called lossy [52], bzip2 [53], and gzip [54]. The experimental results show that all three compressions are not useful for optimized data transfer with bandwidth more than 264.19 Mb/s. This bottleneck imposes the challenge of compression/decompression time reduction to cope with the I/O needs of HEC.
Although compressionbased data reduction methods are feasible for big data reduction, the processing overhead of decompression introduces latency which lowers the performance of analytics algorithms in realtime environments. Moreover, the additional computations consume more cloud resources and increase the overall operational costs of big data systems. However, the techniques enable to store big data efficiently without significantly losing the original data in both the lossy and the lossless compressionbased methods.
3.3 Data Deduplication (Redundancy Elimination)
Data redundancy is the key issue for data analysis in big data environments. Three main reasons for data redundancy are: (1) addition of nodes, (2) expansion of datasets, and (3) data replication. The addition of a single virtual machine (VM) brings around 97% more redundancy, and the growth in large datasets comes with 47% redundant data points [13]. In addition, the storage mechanism for maximum data availability (also called data replication) brings 100% redundancy at the cluster level. Therefore, effective data deduplication and redundancy elimination methods can cope with the challenge of redundancy. The workload analysis shows that the 3× higher throughput improves performance about 45% but in some extreme cases the performance degrades up to 161%. The energy overhead of deduplication is 7%; however, the overall energy saved by processing deduplicated data is 43%. The performance is degraded to 5%, whereas energy overhead is 6% for pure solid state drive (SSD) environments. However, in hybrid environment the system’s performance is improved up to 17%.
Cluster deduplication is a generalized big data reduction scheme for diskbased cluster backup systems. The redundant data stored on multiple disks and partitions are a serious challenge for big data processing systems. The deduplication techniques allow to handle different data chunks (partitions) using hash functions to lower intranode and internode communication overheads. In addition, these methods improve the storage efficiency by eliminating redundant data from multiple nodes. Largescale cluster deduplication schemes face challenge of informationisland (only serverlevel deduplication is possible due to the communication overhead) where data routing is the key issue. Another major challenge is diskchunkindexlookup (keeping duplicated chunk indexes of large datasets creates memory overheads), which degrades the performance of backup clients due to frequently random I/O for lookup and indexing.
In Eq. 13, the \( {\text{CDR}} \) represents the clusterlevel deduplication ratio and the \( {\text{SDR}} \) denotes singlenodelevel deduplication ratio. In addition, \( \alpha \) represents the average usage of storage while \( \sigma \) shows the standard deviation of clusterwide storage usage.
The massive amount of data movement in data centers increases the computational and communicational burdens. The exploitation of innetwork data processing techniques can reduce the aforementioned complexities. The authors of [57] proposed an innetwork data processing technique for bandwidth reduction by customizing routing algorithms, eliminating network redundancy (by caching frequent packets), and reducing onpath data. The proposed technique performs partial data reduction and significantly improved throughput for innetwork query processing.
In contrast, mobile users in the same locality or with the same habits generate similar data points causing a huge amount of redundant data in participatory big data environments. In addition, the absence of spatiotemporal correlation among sensory data in mobile opportunistic networks is also a great challenge. The authors in [58] proposed a cooperative sensing and data forwarding framework for mobile opportunistic networks where sampling redundancy is eliminated to save energy consumption. The authors proposed two data forwarding protocols [epidemic routing with fusion and binary spray and wait with fusion (BSWF)] by leveraging data fusion. The essence of the research is the intelligent fusion of sensory data to eliminate redundancy. The simulation results revealed that proposed method can remove 93% of redundancy in the data as compared to noncooperative methods.
The issue of data duplication or redundancy has been addressed by researchers in different environments at different levels (mobile, cluster, cloud, and data center). Therefore, the selection of best method depends upon the application models. For example, in mobile opportunistic networks and mobile crowd sensing environments, the data redundancy elimination methods are best suited when they are deployed in mobile devices. Similarly, for scientific and highly correlated data deduplication is best suitable when it is performed at cluster, data center, and cloud level.
3.4 Data Preprocessing
Data preprocessing is the second important phase of big data processing, and it must be preprocessed before storage at largescale infrastructures [19]. This approach helps in big data reduction and also extracts the metadata for further processing. The authors argue that primary approaches for data preprocessing are based on semantic analysis (using ontologies) or linked data structures (such as Google knowledge graph). However, this literature review uncovers few other techniques, such as low memory prefilters for streaming data, URL filtration method, and mapreduce implementation of 2D peak detection methods in the big genomic data.
Low memory prefilters are used for preprocessing genomic sequencing streaming data. The algorithm runs in a single pass and gives improved performance for error correction and lossy compression in data streams. In addition, the algorithm extracts the subsets of data streams using sketchbased techniques and applies prefiltering algorithm for lossy compression and error correction. The algorithm first constructs the Bruijn graph, and the subsets are extracted using locusspecific graph analysis technique [59]. The massive data redundancy is handled using the \( k \)mers median, and subsequently, digital normalization is employed as the data reduction technique. The authors argued that 95% of the data can be removed in the normal sequencing sample and the percentage reaches 98% of highcoverage single sequencing data. The results show that memory requirement for proposed algorithm is reduced from 3 TB to 300 GB of RAM.
Wearable sensors generate multidimensional, nonlinear, dynamic data streams with weak correlation between data points. The authors in [60] used localitysensitive bloom filter to enhance the efficiency of instancebased learning for frontend data preprocessing near the sensing elements. The technique enables the filtration and communication of only the relevant and meaningful information to reduce computational and communication burden. The authors discussed the big healthcare data system for elderly patients and developed a prototype of the proposed solution. The architecture of the system is based upon a wearable sensor with bluetooth low energy (BLE) interface and can communicate with mobile application and/or PC to establish a personal area network (PAN). The mobile application processes the data and recognizes the state of the user. The sensor data and user states are further transmitted to a remote big data server through TCP/UDP ports. The compression algorithms are applied to incoming data streams, and resultant compressed files remain 10% of the actual data streams.
An application of big data reduction is the filtration of malicious URLs in Internet security applications. The authors in [21] proposed two feature reduction techniques that extract the lexical features and the descriptive features and then combine their results. The lexical features extract the structure of the URLs. However, the issue with lexical features is that malicious URL addresses have constantly changing behavior to abstain from malware detection software. The descriptive features are extracted to track and preserve different states of the same URL to label it as malicious. The authors selected passiveaggressive (for dense feature extraction) and confidence weighted algorithms (for sparse feature extraction) as the online learning algorithms and trained their models with extracted features [61, 62]. The prediction results of the filtration technique demonstrate around 75% data reduction with approximately 90% retention rate (inverse of data loss).
The analysis of largescale proteomics data, which is the proteinlevel representation of big genomic data, requires massive computational resources to study different protein properties, such as expressions, changes in protein structures, and the interaction with other proteins. The protein molecules are too large to be identified by spectrometer and therefore are broken into smaller fragments called peptides. The mass spectrometer outputs the graphical output where each spectrum of data points is shown using Gaussian curves for peptide identification. The preprocessing step of proteomics data analysis is the identification of curves also called the 2D peaks. Each of the samples submitted to the spectrometer takes around 100 min to 4 h for complete analysis. During the passage of peptides, the spectrometer takes snapshots of spectrum every second where each peptide remains visible for several spectrums. The authors proposed a mapreduce implementations for proteomics data analysis where 2D peaks are picked at map level and further analyzed at reduce level [63]. The data reduction takes place at map level by applying preprocessing techniques for decoding the arrays, noise removal, and management of the overlapping peaks in the spectrum. Experimental results show that the given mapreduce implementation completes the data analysis in 22 min.
Recently light detection and ranging (LiDAR) technology enabled the generation of big 3D spatial data [64]. A cloud computingbased LiDAR processing system (CLiPS) processes big 3D spatial data effectively. The CLiPS uses several preprocessing techniques for data reduction to deal with large size of data. The data reduction is performed using a vertex decimation approach to provide a user’s preferred parameters to reduce the big data. The results show the advantage of cloud computing technologies over the conventional systems comparing performance and time consumption.
The literature review of these techniques reveals that data preprocessing techniques are highly dependent on the nature of big data and also encourage further investigation of the underlying problem. Therefore, these techniques could not be generalized for all types of big data streams.
3.5 Dimension Reduction
Dimension reduction methods
References  Methods  Description  Strengths  Weaknesses 

Weinstein et al. [8]  DQC  Visual data mining method  Ability to expose hidden structures and determine their significance in highdimensional big data  Lacks efficiency Requires a combination of statistical tools 
Hsieh et al. [9]  BIGQuic  Applying a parallel divideandconquer strategy  Supports parallelization Allowing for inexact computation of specific components  Lacks accuracy and reliability 
Hoi et al. [68]  OFS  Selection of active features for accurate prediction  Works best in online learning mode as compared with batchmode learning  Lacks efficiency 
Feldman et al. [11]  Corsets  Applying corsets to reduce big data  High significance when used for data complexity  Works well on small datasets only 
Azar and Hassanien [71]  LHNFCSF  Linguistic hedges fuzzy classifier  Data reduction  Lack of efficiency 
Cichocki [72]  TNs  Tensor decomposition and approximation  Reduction in feature spaces  High computational complexity 
Dalessandro [73]  FH  Maps features from highdimensional space to lowdimensional spaces  Reduces feature space randomly  Compromise on data quality 
Liu et al. [77]  CF  Classifier training with minimal feature spaces  Outlines critical feature dimensions and adequate sampling size  Assumptions need to be more accurate to outline critical feature dimension 
Zeng and Li [74]  IPLS  Performs incremental analysis of streaming data  Computationally efficient Highly accurate  Needs to handle change detection in streaming data 
Conventional dimensionality reduction algorithms that use Gaussian maximum likelihood estimator could not handle the datasets with over 20,000 variables. The BIGQuic addresses the issue by applying a parallel divideandconquer strategy that can be applied up to 1million variables in the feature space for dimensionality reduction [9]. The results show that the proposed algorithm is highly scalable and faster than the existing algorithms, such as Glasso and ALM [65, 66].
Normally, online learning techniques take the full feature set as the input, which is quite challenging when dealing with highdimensional features space. The authors proposed an online feature selection (OFS) approach where the online learners only work on small and fixedlength feature sets. However, the selection of active features for accurate prediction is a key issue in the approaches presented in [68]. The authors investigated sparsity regularization and truncation techniques and proposed a new algorithm called the OFS. The results showed that the OFS outperformed RAND and \( {\text{PE}}_{\text{trun}} \) algorithms for UCI datasets and it works best in online learning mode as compared to batchmode learning.
The corsets are the small set of points that represent the larger population of data and maintain the actual properties of overall population. These properties vary by nature of knowledge discovery algorithms. For example, the corsets representing first kcomponents maps with first kcomponents in the big data. Similarly, the corsets containing kclusters with radius r approximate the big data and obtain the kclusters with same r. In this way, the authors [11] applied corsets to reduce the big data into small and manageable size, which reduces the overall data complexity. The authors mapped corsets with kmeans, principal component analysis (PCA) and projective clustering algorithms deployed with massively parallel streaming data [69, 70]. The big data is reduced in such a way that high dimensions of input space do not affect the cardinalities of corsets. In addition, the corsets are merged by maintaining the property that union of two corsets represents the reduced set of union of two big datasets. The experimental results showed that corsets are suitable to address NPhard problems in massively parallel and streaming data environments where big data complexity is reduced by application of data processing algorithms on small datasets that are approximated from big data. In addition, the corsets are paradigm shift in big data analysis where the focus of research remains on big data reduction instead of improving the computational efficiency of existing algorithms.
Medical big data comes across several issues regarding extraction of structures, storage of massive data streams, and uncovering the useful knowledge patterns. Research shows that fuzzy classification methods are good choice to cope with the abovementioned issues. Recently, the authors of [71] presented linguistic hedges fuzzy classifier with selected features (LHNFCSFs) to reduce dimensions, select features, and perform classification operations. The integration of linguistic hedges in adaptive neuralfuzzy classifier enhances the accuracy. The LHNFCSF reduces the feature space effectively and enhances the performance of the classifier by removing unimportant, noisy, or redundant features. The results depict that the LHNFCSF addresses the medical big data issues by reducing the dimensions of large datasets and speeding up the learning process and improves the classification performance.
Tensors are multidimensional representations of data elements with at least one extra dimension as compared to matrices. The increasing numbers of elements demand more computational power to process the tensors. Tensor processing works fine with small tensors. However, processing large tensors is a challenging task [10]. Tensor decomposition (TD) schemes are used to extract small but representative tensors from large tensors [72]. Three widely used TD strategies include canonical polyadic decomposition (CPD), tucker decomposition, and tensor trains (TT). The TD schemes represent the large tensors linked with their small representations. These decomposition schemes reduce the high dimensionality in big datasets and establish the interconnection among tensors to form tensor networks (TNs). These TNs enable to further reduce the data size by using optimizationbased algorithms to find factor matrices and optimize using linear and nonlinear least square methods. The case studies show that tensor decomposition strategies could be used to alleviate/eliminate dimensionality in large scientific computing datasets and have many potential applications for feature extraction, cluster analysis, classification, data fusion, anomaly detection, pattern recognition, integration, timeseries analysis, predictive modeling, multiway component analysis, and regression.
The feature hashing (FH) method reduces feature dimensionality by randomly assigning each feature in the actual space to a new dimension in a lowerdimensional space [73]. This is done by simply hashing the ID of the original features. Usually, all dimensional reduction techniques degrade the data quality. However, most of them preserve the geometric qualities of the data. Alternately, the FH does not preserve the data quality. Research shows that the degradation of data quality is so minimal that its benefits are outweighed by the cost. The FH scales linearly with simple preprocessing and preservation of data sparsity, if exists. The scalability property of the FH makes it a natural choice for million (or even billion) feature datasets. For example, the FH method applied to email spam filtering shows its power when applied upon sparse and streaming data with realtime requirements of mass customization. The results show that the feature set is reduced from one billion to one million features.
Big data streams enter with episodic information and create highdimensional feature spaces. Normally, the feature extraction methods need whole data in the memory that increases the computational complexity of big data systems and degrades the performance of classifiers. The incremental feature extraction methods are the best choice to handle such issues [74]. Incremental partial least squares (IPLSs) is a variant of the partial least squares method that effectively reduces dimensions from largescale streaming data and improves the classification accuracy. The proposed algorithm works in twostage feature extraction process. First, the IPLS adopts the target function to update the historical means and to extract the leading projection direction. Second, the IPLS calculates the rest of projection directions that are based on the equivalence between the Krylov sequence and the partial least square vectors [75]. The comparison of the IPLS was performed with incremental PCA algorithm, incremental interclass scatter method, and incremental maximum margin criterion technique. The results revealed that the IPLS showed improved performance in terms of accuracy and computational efficiency.
Systems of systems (SoS)—case study The integration of heterogeneous and independent operating computing systems to collectively maximize the performance as compared to the individual settings leads toward the SoS [76]. Nowadays, SoS is contributing to generate big data and raises the need for data reduction. Few examples of statistical and computational intelligence tools for data reduction in SoS include the PCA, clustering, fuzzylogic, neurocomputing, and evolutionary computing, such as genetic algorithms, and Bayesian networks. The authors applied data reduction methods at different stages of analyzing photovoltaic data that were collected from different sources. The original dataset contained 250 variables, which is highly dimensional and is not practical due to limitations of execution time and memory constraints on a desktop computer. Two approaches for data reduction at this stage were considered: (1) labeling the interesting attributes by domain expert and (2) development of an adaptive learning algorithm for automatic attribute selection. The authors employed the first approach for data reduction. The authors further cleanedup the data and removed all invalid data points from the dataset. For example, solar irradiance in night hours generates data points with negative values, therefore not feasible for contribution in the study. After removing the invalid data, the data points containing very low values for global horizontal irradiance (GHI), direct normal irradiance (DNI), and direct horizontal irradiance (DHI) are removed to create more crispy data for further analysis.
The cleaned data are further fed into two nonparametric model generation tools, namely the fuzzy inference system generator and backpropagation neural network training tools using MATLAB fuzzy logic toolbox and the neural network toolbox. The initial evaluation of both of the tools revealed that the input variables should be further reduced for performance maximization in terms of execution time and memory consumption. The authors expanded the nonlinear data by using additional variables that in turn increased the performance of the training model but also increased time and space complexity. Therefore, the PCA is applied for dimension reduction to compress the data without significant information loss. After application of the PCA, further dimension reduction was performed using genetic algorithm (GA). First, the data were reduced using the GA on the full set of initial data and remaining data were expanded nonlinearly. Finally, the expanded dataset is used to train a neural network to assess the overall effectiveness of the GA. In practice, the time and computationrelated constraints were limited to the selection of training data to first 1000 samples. The first iteration of GA took initially 244 samples and reduced it to 74. The results showed that the GA reduced the number of attributes up to 70%.
3.6 Data Mining and Machine Learning (DM and ML)
Recently, several DM and ML methods have also been proposed for big data reduction. The methods are either applied to reduce data immediately after its acquisition or customized to address some specific problems. For example, the authors [78] proposed a contextaware big data forwarding scheme using distributed wearable sensors. The scheme is based on hidden markov model (HMM) to enable contextaware features in the distributed sensor model and forwards only relevant information when there is some significant change in the data [78]. The proposed scheme reduces the communication and storage overhead in big data environment. The authors compared the result of proposed localitysensitive hash (LSH)based method with linear perception (LP)based dimensionality reduction algorithm and argued on the effectiveness of the proposed scheme as compared to the LPbased dimensionality reduction methods.
The problem of mining uncertain big data due to existential probabilities becomes worse and requires huge efforts and computational power to explore the incrementally growing uncertain search space. Therefore, the search space is needed to be reduced for uncovering the maximum certain and useful patterns. The authors of [79] proposed a mapreduce algorithm to reduce the search space and mine frequent patterns from uncertain big data. The algorithm facilitates the users to confine their search space by setting some succinct antimonotone (SAM) constraints for data analysis and subsequently mines the uncertain big data to uncover frequent patterns that satisfy the userspecified SAM constraints. The input of the algorithm is uncertain big data, userspecified minimum support (minsup) threshold, and the SAM constraints. Two sets of mapreduce implementations are used to uncover singleton and nonsingleton frequent patterns. Experimental results show that the userspecified SAM (termed as selectivity) is directly proportional to multiple parameters which are derived from algorithm’s runtime, the pairs returned by map function, the pairs sorted and shuffled by reduce function, and the required constraints checks.
Artificial intelligence methods, for example, artificial neural networks (ANNs) have also potential for big data reduction and compression. The authors in [80] proposed a selforganized Kohonen networkbased method to reduce big hydrographic data acquired from the deep seas. The proposed system first converts the raw data into ‘swath’ files using a combination of four filters: (1) limit filter, (2) amplitude filter, (3) along track filter, and (4) across track filter. The appropriate combinations of the filters ensure the optimal dataset size. Despite filtering, the sample size is still large to be considered as big data. The selforganized Kohonen networks are trained and optimized using filtered hydrographic data to cluster the incoming data streams. The experimental results exhibited the feasibility of selforganized Kohonen networks for big hydrographic data for further analysis.
Deep learning models are inherently computationally complex and require manycore CPUs and largescale computational infrastructures. Some recent learning approaches for such largevolume, complex data include locally connected networks [82, 83], improved optimizers, and deep stacking networks (DSNs). The authors of [84] proposed the hybrid deep learning model, called DisBelief, to address the issue of high dimensionality in big data streams. Disbelief utilizes a largescale cluster to parallelize both the data and the learning models using synchronization, message passing, and multithreading techniques. The DisBelief model first achieves parallelism by partitioning largescale networks into small blocks that are mapped to a single node and then achieves data parallelism using two separate distribution optimization procedures called stochastic gradient descent (SGD) for online optimization and sandblaster for batch optimization. Although feasible for big data reduction, the deep learning models are resource hungry and require MPIs based on clusters of CPUs or GPUs. Therefore, there is a need to develop optimized deep learning strategies to achieve resource efficiency and reduce communication burdens inside the MPIs.
The wide spectrum view of the proposed methods for big data reduction uncovers the fact that the research on big data reduction methods is being carried out at several levels of big data architecture and in different forms.
4 Open Research Issues
The discussion on the open research issues, limitations, and possible future research directions is presented in this section.
Network theory The extraction of topological network and ranking of network nodes from big data is a complex process due to inherent big data complexity. In addition, the complex interactions among different nodes of the extracted networks increase the computational complexity of existing network theorybased methods. The scalefree networks and random networks can effectively reduce complex big datasets. However, the full network extraction from inconsistent and missing data is the key challenge [16, 40]. Big data systems contain many small and manageable datasets, but finding the connections among these datasets is a crucial task. The similarity graph is generated from big data where vertices represent datasets and the weighted edges are defined on the basis of similarity measure. The graph is further reduced by merging similar datasets to reduce the number of nodes. The similaritybased big data reduction methods are good choice for network extraction and reduction. However, a range of new similarity measures are required to deal with the evolving complexity and to fully comply with 6Vs of big data [17].
Persistent homology is a good solution for topological data analysis, but it involves high computational complexity. The solutions like selective collapse algorithms represent datasets in the form of forests, and the nodes are collapsed in a way to improve the speed of persistent homology and maintain strong collapse. The persistent homology tools for reducing and analyzing big data still need to be further explored in the future research [18, 42]. Similarly, the automated extraction of events and their representation in network structures is an emerging research area. The assessment of events cooccurrence and their mutual influences is the key challenge for big data reduction. The authors in [41] performed the influence assessment among different concepts (events or datasets) based on the cooccurrence of two events. The cooccurrence is assessed based on the preferential attachment property which determines that new nodes are most likely connected with highly connected nodes as compared to less connected nodes. In addition, the influence relationship among network nodes can be effectively derived from conditional dependencies among variables. However, the mathematical and probabilistic constraints increase the computational complexity in network extraction methods. Therefore, efforts are required to optimize the influence assessment methods for computationally efficient and better approximated network structures [41].
Compression Big data processing in cloud computing environments involves challenges relevant to inefficiency, parallel memory bottlenecks, and deadlocks. The spatiotemporal compression is a key solution for processing big graph data in the cloud environment. In spatiotemporal compressionbased methods, the graph is partitioned and edges are mapped into different clusters where compression operations are performed for data reduction. The spatiotemporal compression is an effective approach for big data reduction. However, the research is required to find new parameters that are helpful in finding additional spatiotemporal correlations for maximum big data reduction [45].
The gap between computations and I/O capacity in the HEC systems degrades the system performance significantly. Although in situ analytics are useful for decreasing the aforementioned gap, the cost of computation increases abruptly. The compression methods can significantly reduce the transferred data and narrow the gap between computations and I/O capacity. The authors in [43] suggested that the number of available processors and the data reduction ratio (compression ratio) are two key factors that need attention in future research in this area. Alternately, the AST is a new way of compressing digitized data by selectively stretching and warping the signal. The technique is primarily based on selfadaptive stretch where more samples are associated with sharp features and fewer samples are associated with redundant coarse features. The AST performs data compression of the signal extracted on frequency domain. The method also performs inverse transformation of the constructed signal. The method specifically works with big data involving signal processing. However, the generalization to other domains is a bottleneck in this research [47].
Compressed sensing is a compressible and/or sparse signal that projects a highdimensional data in lowdimensional space using random measurement matrix. The proposed scheme performs data acquisition and compression in parallel for improved performance as compared with Nyquist sampling theorybased compression methods. The probability of poor data quality and information fidelity loss increases when the analysis is performed on reduced and compressed data [48]. The RED encoding scheme proposed by authors in [48] is used to manage massively generated voluminous electrophysiology data. The scheme performs best when encoding of invariant signals is performed. However, while encoding timeseries signals, the performance varies but the scheme achieves high compression rate with improved computational speed in lossless compression. The performance of the RED encoding methods degrades with high variance in signals [50].
Parallel compression methods can be used to reduce the data size with low computational cost. It uses proper orthogonal decomposition to compress data because it can effectively extract important features from the data and resulting compressed data can also be linearly decompressed. The parallel compression methods balance between feature retention error and compression ratio and perform fast decompression for interactive data visualization. However, the standard deviation of error is significant due to noise in the dataset [55]. The sketching method uses countmin sketch algorithm to compress vehicular movement data and achieve compact communication. Although it ensures data reduction by preserving some important characteristics of the original data, the probability of information fidelity loss is more when sketching is applied with inconsistent and noisy data stream [46].
Data deduplication (redundancy elimination) Clusterlevel data deduplication is a key requirement to comply with servicelevel agreements (SLAs) for privacy preserving in cloud environments. The main challenge is the establishment of tradeoff between high deduplication ratio and scalable deduplication throughput. The similaritybased deduplication scheme optimizes the elimination process by considering the locality and similarity of data points in both the intranode and internode scenarios. The approach is effective for data reduction, but it requires to be implemented with very largescale cluster data deduplication systems [12]. The I/O latency and extra computational overhead of clusterlevel data deduplication are among the key challenges. The authors in [13] characterized the deduplication schemes in terms of energy impact and performance overhead. The authors outlined three sources of redundancy in cluster environment including: (1) the deployment of additional nodes in the cluster, (2) the expansion of big datasets, and (3) the usage of replication mechanisms. The outcomes of the analysis reveal that the local deduplication, at cluster level, can reduce the hashing overhead. However, local deduplication cannot achieve the maximum redundancy. In contrast, global deduplication can achieve maximum redundancy but compromises on the hashing overheads. In addition, finegrained deduplication is not suitable for big datasets especially in streaming data environments [13].
Data routing is a key issue in multinode data deduplication systems. The availability of sufficient throughput is the main bottleneck for data movement among backup and recovery systems. The stateful data routing schemes, as compared to stateless approaches, have higher overhead with low imbalance in the data which minimizes the utility of data deduplication systems. The open issues for data routing include the characterization of parameters which causes the data skew. In addition, the scalability of routing methods to largescale cluster systems and the impact of feature selection and superchunk size are needed to be explored in future research. Moreover, the addition of new nodes is needed to be considered for effective bin migration strategies [14].
The innetwork data processing methods facilitate in data reduction and reduce the bandwidth consumption, and the efforts are required for onthepath data reduction and redundancy elimination. The reduced bandwidth consumption by innetwork data processing methods enable enhanced query processing throughput. The future implementation of innetwork data processing is envisioned as the provision of networkasaservice (NaaS) in the cloud environment which is fully orchestrated for redundancy elimination and query optimization [57]. In addition, there is a need to devise new networkaware query processing and optimization models, and integration of these models in distributed data processing systems. Research shows that cooperative sensing methods can aid in significant data reduction in largescale sensing infrastructures [58]. Current cooperative sensing methods lack in lowlevel contextual features and adaptive global learning models to handle the change detection in streaming data environments. Future research work to integrate current lowlevel contextual models and adaptive machine learning methods can aid in maximum data reduction as well as collection of a highquality data.
Data preprocessing The investigations of research problems relevant to preprocessing techniques of big data are still at the initial level. Most of the works are based on the adoption of existing preprocessing methods that were earlier proposed for historical large datasets and data streams. The forefront deployment of data preprocessing methods in the big data knowledge discovery process requires new, efficient, robust, scalable, and optimized preprocessing techniques for both historical and streaming big data. The application of appropriate and highly relevant preprocessing methods not only increases data quality but also improves the analytics on reduced datasets. The research on new methods for sketching, anomaly detection, noise removal, feature extraction, outliers detection, and prefiltering of streaming data is required to reduce big data effectively. In addition, the deployment of adaptive learning models in conjunction with said methods can aid in dynamic preprocessing of big streaming data [21].
Dimension reduction Big data reduction is traditionally considered to be a dimension reduction problem where multimillion features spaces are reduced to manageable feature spaces for effective data management and analytics. Unsupervised learning methods are the key consideration for dimensionality reduction problem. However, this literature review revealed several other statistical and machine learning methods to address this issue. The techniques to combine conventional dimension reduction methods with statistical analysis methods can increase the efficiency of big data systems [8]. This approach may aid in targeting highly dense and information oriented structures (feature sets) to achieve maximum and efficient big data reduction. Alternately, tensor decomposition and approximation methods are useful to cope with the curse of dimensionality that arises due to highdimensional complex and sparse feature spaces [10]. The main application of TDbased methods is witnessed in the scientific computing and quantum information theory domain. This literature review revealed that the issue of dimensionality reduction in big data could be handled by adopting frontend data processing, online feature selection from big data streams, constantsize corsets for clustering, statistical methods, and fuzzy classificationbased soft computing approaches. These adoptions open new research avenues for interdisciplinary research and develop novel big data reduction methods. The strengths and weaknesses of these methods are already presented in detail in Table 2.
DM and ML The DM and the ML methods for big data reduction could be used at various levels of big data architectures. These methods enable to find interesting knowledge patterns from big data streams to produce highly relevant and reduced data for further analysis. For example, HMM as applied in [78] enables the contextaware features to filter the raw data streams to transmit only highly relevant and required information. In addition, the scheme enables to project highdimensional data streams in manageable lowdimensional feature spaces. Although the application of these methods is convenient for data reduction, the tradeoff between energy consumptions in local processing with raw data transmission is a key challenge that is needed to be considered. The DM and ML methods also have potential to be deployed in mapreduce implementations of Hadoop architecture. The authors in [79] parallelized the frequent pattern mining algorithms using the mapreduce programming model to reduce the massively highdimensional feature space produced by uncertain big data. However, there exists a huge research gap for the implementation of other DM and ML methods for big data reduction that include supervised, unsupervised, semisupervised, and hierarchical deep learning models [85]. In addition, the implementation of statistical methods, both descriptive and inferential, for big data reduction using approximation and estimation properties in uncertain big data environments is also useful for data reduction in mapreduce programming models. Moreover, the DM and ML methods are equally useful for big data reduction when coupled with artificial intelligencebased optimization methods. However, supervised, unsupervised, and semisupervised learning methods need more attention for future research [80].
Deep learning models have recently gained attention by the researchers. The deployment of deep learning models for big data reduction is potential research direction that can be pursued in future. The deep learning models are initially developed from certain data and gradually evolve with uncertain data to effectively reduce big data streams. However, the increasing computational complexities of operating in uncertain big data environments and optimization of learning models to discover patterns from maximum data are the issues that can be further investigated [84].
In this section, we thoroughly discussed the open issues, research challenges, the limitations of proposed methods for big data reduction and presented some future research directions. The survey reveals that big data reduction is performed at many levels during the data processing lifecycle that include data capturing, data preprocessing, data indexing and storage, data analysis, and visualization. Therefore, the relevant reduction methods and systems should be designed to handle the big data complexity at all stages of big data processing. In addition, the future research work should focus on considering all 6Vs to process big data in computing systems with different form factors from finegrained mobile computing systems to largescale massively parallel computing infrastructures.
5 Conclusions
Big data complexity is a key issue that is needed to be mitigated. The methods discussed in this article are an effort to address the issue. The presented literature review reveals that there is no existing method that can handle the issue of big data complexity singlehandedly by considering the all 6Vs of big data. The studies discussed in this article mainly focused on data reduction in terms of volume (by reducing size) and variety (by reducing number of features or dimensions). However, further efforts are required to reduce the big data streams in terms of velocity and veracity. In addition, the new methods are required to reduce big data streams at the earliest immediately after data production and its entrance into the big data systems. In general, compressionbased data reduction methods are convenient for reducing volume. However, the decompression overhead needs to be considered to improve efficiency. Similarly, network theorybased methods are effective for extracting structures from unstructured data and to efficiently handle the variety in big data. The data deduplication methods are useful to improve the data consistency. Therefore, the aforementioned methods are a suitable alternative to manage the variability issues in big data. Likewise, data preprocessing, dimension reduction, data mining, and machine learning methods are useful for data reduction at different levels in big data systems. Keeping in view the outcomes of this survey, we conclude that big data reduction methods are emerging research area that needs attention by the researchers.
References
 1.Wu X et al (2014) Data mining with big data. IEEE Trans Knowl Data Eng 26(1):97–107CrossRefGoogle Scholar
 2.Che D, Safran M, Peng Z (2013) From big data to big data mining: challenges, issues, and opportunities. In: Database systems for advanced applicationsGoogle Scholar
 3.Battams K (2014) Stream processing for solar physics: applications and implications for big solar data. arXiv preprint arXiv:1409.8166
 4.Zhai Y, Ong YS, Tsang IW (2014) The emerging “big dimensionality”. Comput Intell Mag IEEE 9(3):14–26CrossRefGoogle Scholar
 5.Fan J, Han F, Liu H (2014) Challenges of big data analysis. Nat Sci Rev 1(2):293–314CrossRefGoogle Scholar
 6.Chandramouli B, Goldstein J, Duan S (2012) Temporal analytics on big data for web advertising. In: 2012 IEEE 28th international conference on data engineering (ICDE)Google Scholar
 7.Ward RM et al (2013) Big data challenges and opportunities in highthroughput sequencing. Syst Biomed 1(1):29–34CrossRefGoogle Scholar
 8.Weinstein M et al (2013) Analyzing big data with dynamic quantum clustering. arXiv preprint arXiv:1310.2700
 9.Hsieh CJ et al (2013) BIG & QUIC: sparse inverse covariance estimation for a million variables. In: Advances in neural information processing systemsGoogle Scholar
 10.Vervliet N et al (2014) Breaking the curse of dimensionality using decompositions of incomplete tensors: tensorbased scientific computing in big data analysis. IEEE Signal Process Mag 31(5):71–79CrossRefGoogle Scholar
 11.Feldman D, Schmidt M, Sohler C (2013) Turning big data into tiny data: constantsize coresets for kmeans, pca and projective clustering. In: Proceedings of the twentyfourth annual ACMSIAM symposium on discrete algorithmsGoogle Scholar
 12.Fu Y, Jiang H, Xiao N (2012) A scalable inline cluster deduplication framework for big data protection. In: Middleware 2012. Springer, pp 354–373Google Scholar
 13.Zhou R, Liu M, Li T (2013) Characterizing the efficiency of data deduplication for big data storage management. In: 2013 IEEE international symposium on workload characterization (IISWC)Google Scholar
 14.Dong W et al (2011) Tradeoffs in scalable data routing for deduplication clusters. In: FASTGoogle Scholar
 15.Xia W et al (2011) SiLo: a similaritylocality based nearexact deduplication scheme with low RAM overhead and high throughput. In: USENIX annual technical conferenceGoogle Scholar
 16.Trovati M, Asimakopoulou E, Bessis N (2014) An analytical tool to map big data to networks with reduced topologies. In: 2014 international conference on intelligent networking and collaborative systems (INCoS)Google Scholar
 17.Fang X, Zhan J, Koceja N (2013) Towards network reduction on big data. In: 2013 international conference on social computing (SocialCom)Google Scholar
 18.Wilkerson AC, Chintakunta H, Krim H (2014) Computing persistent features in big data: a distributed dimension reduction approach. In: 2014 IEEE international conference on acoustics, speech and signal processing (ICASSP)Google Scholar
 19.Di Martino B et al (2014) Big data (lost) in the cloud. Int J Big Data Intell 1(1–2):3–17CrossRefGoogle Scholar
 20.Brown CT (2012) BIGDATA: small: DA: DCM: lowmemory streaming prefilters for biological sequencing dataGoogle Scholar
 21.Lin MS et al (2013) Malicious URL filtering—a big data application. In 2013 IEEE international conference on big dataGoogle Scholar
 22.Chen J et al (2013) Big data challenge: a data management perspective. Front Comput Sci 7(2):157–164MathSciNetCrossRefGoogle Scholar
 23.Chen XW, Lin X (2014) Big data deep learning: challenges and perspectives. IEEE Access 2:514–525CrossRefGoogle Scholar
 24.Chen Z et al (2015) A survey of bitmap index compression algorithms for big data. Tsinghua Sci Technol 20(1):100–115MathSciNetCrossRefGoogle Scholar
 25.Hashem IAT et al (2015) The rise of “big data” on cloud computing: review and open research issues. Inf Syst 47:98–115CrossRefGoogle Scholar
 26.Gani A et al (2015) A survey on indexing techniques for big data: taxonomy and performance evaluation. In: Knowledge and information systems, pp 1–44Google Scholar
 27.Kambatla K et al (2014) Trends in big data analytics. J Parallel Distrib Comput 74(7):2561–2573CrossRefGoogle Scholar
 28.Jin X et al (2015) Significance and challenges of big data research. Big Data Res 2(2):59–64CrossRefGoogle Scholar
 29.Li F, Nath S (2014) Scalable data summarization on big data. Distrib Parallel Databases 32(3):313–314CrossRefGoogle Scholar
 30.Lohr S (2014) For bigdata scientists, ‘janitor work’ is key hurdle to insights. http://www.nytimes.com/2014/08/18/technology/forbigdatascientistshurdletoinsightsisjanitorwork.html
 31.Ma C, Zhang HH, Wang X (2014) Machine learning for big data analytics in plants. Trends Plant Sci 19(12):798–808CrossRefGoogle Scholar
 32.Ordonez C (2013) Can we analyze big data inside a DBMS? In: Proceedings of the sixteenth international workshop on data warehousing and OLAPGoogle Scholar
 33.Oliveira J, Osvaldo N et al (2014) Where chemical sensors may assist in clinical diagnosis exploring “big data”. Chem Lett 43(11):1672–1679CrossRefGoogle Scholar
 34.Shilton K (2012) Participatory personal data: an emerging research challenge for the information sciences. J Am Soc Inform Sci Technol 63(10):1905–1915CrossRefGoogle Scholar
 35.Shuja J et al (2012) Energyefficient data centers. Computing 94(12):973–994CrossRefzbMATHGoogle Scholar
 36.Ahmad RW et al (2015) A survey on virtual machine migration and server consolidation frameworks for cloud data centers. J Netw Comput Appl 52:11–25CrossRefGoogle Scholar
 37.Bonomi F et al (2014) Fog computing: a platform for internet of things and analytics. In: Big data and internet of things: a roadmap for smart environments. Springer, pp 169–186Google Scholar
 38.Rehman MH, Liew CS, Wah TY (2014) UniMiner: towards a unified framework for data mining. In: 2014 fourth world congress on information and communication technologies (WICT)Google Scholar
 39.Patty JW, Penn EM (2015) Analyzing big data: social choice and measurement. Polit Sci Polit 48(01):95–101CrossRefGoogle Scholar
 40.Trovati M (2015) Reduced topologically realworld networks: a bigdata approach. Int J Distrib Syst Technol (IJDST) 6(2):13–27CrossRefGoogle Scholar
 41.Trovati M, Bessis N (2015) An influence assessment method based on cooccurrence for topologically reduced big data sets. In: Soft computing, pp 1–10Google Scholar
 42.Dey TK, Fan F, Wang Y (2014) Computing topological persistence for simplicial maps. In: Proceedings of the thirtieth annual symposium on computational geometryGoogle Scholar
 43.Zou H et al (2014) Flexanalytics: a flexible data analytics framework for big data applications with I/O performance improvement. Big Data Res 1:4–13CrossRefGoogle Scholar
 44.Ackermann K, Angus SD (2014) A resource efficient big data analysis method for the social sciences: the case of global IP activity. Procedia Comput Sci 29:2360–2369CrossRefGoogle Scholar
 45.Yang C et al (2014) A spatiotemporal compression based approach for efficient big data processing on Cloud. J Comput Syst Sci 80(8):1563–1583MathSciNetCrossRefzbMATHGoogle Scholar
 46.Monreale A et al (2013) Privacypreserving distributed movement data aggregation. In: Geographic information science at the heart of Europe. Springer, pp 225–245Google Scholar
 47.Jalali B, Asghari MH (2014) The anamorphic stretch transform: putting the squeeze on “big data”. Opt Photonics News 25(2):24–31CrossRefGoogle Scholar
 48.Wang W et al (2013) Statistical waveletbased anomaly detection in big data with compressive sensing. EURASIP J Wirel Commun Netw 2013(1):1–6CrossRefGoogle Scholar
 49.He B, Li Y (2014) Big data reduction and optimization in sensor monitoring network. J Appl Math. doi: 10.1155/2014/294591
 50.Brinkmann BH et al (2009) Largescale electrophysiology: acquisition, compression, encryption, and storage of big data. J Neurosci Methods 180(1):185–192CrossRefGoogle Scholar
 51.Zou H et al (2014) Improving I/O performance with adaptive data compression for big data applications. In: 2014 IEEE international parallel & distributed processing symposium workshops (IPDPSW)Google Scholar
 52.Lakshminarasimhan S et al (2011) Compressing the incompressible with ISABELA: in situ reduction of spatiotemporal data. In: EuroPar 2011 parallel processing. Springer, pp 366–379Google Scholar
 53.Ahrens JP et al (2009) Interactive remote largescale data visualization via prioritized multiresolution streaming. In: Proceedings of the 2009 workshop on ultrascale visualizationGoogle Scholar
 54.Compression utility, gzip. http://www.gzip.org
 55.Bi C et al (2013) Proper orthogonal decomposition based parallel compression for visualizing big data on the K computer. In: 2013 IEEE symposium on largescale data analysis and visualization (LDAV)Google Scholar
 56.Bhagwat D, Eshghi K, Mehra P (2007) Contentbased document routing and index partitioning for scalable similaritybased searches in a large corpus. In: Proceedings of the 13th ACM SIGKDD international conference on knowledge discovery and data miningGoogle Scholar
 57.Rupprecht L (2013) Exploiting innetwork processing for big data management. In: Proceedings of the 2013 SIGMOD/PODS Ph.D. symposiumGoogle Scholar
 58.Zhao D et al (2015) COUPON: a cooperative framework for building sensing maps in mobile opportunistic networks. IEEE Trans Parallel Distrib Syst 26(2):392–402CrossRefGoogle Scholar
 59.Zerbino DR, Birney E (2008) Velvet: algorithms for de novo short read assembly using de Bruijn graphs. Genome Res 18(5):821–829CrossRefGoogle Scholar
 60.Cheng Y, Jiang P, Peng Y (2014) Increasing big data front end processing efficiency via locality sensitive Bloom filter for elderly healthcare. In: 2014 IEEE symposium on computational intelligence in big data (CIBD)Google Scholar
 61.Dredze M, Crammer K, Pereira F (2008) Confidenceweighted linear classification. In: Proceedings of the 25th international conference on machine learningGoogle Scholar
 62.Crammer K et al (2006) Online passiveaggressive algorithms. J Mach Learn Res 7:551–585MathSciNetzbMATHGoogle Scholar
 63.Hillman C et al (2014) Near realtime processing of proteomics data using Hadoop. Big Data 2(1):44–49MathSciNetCrossRefGoogle Scholar
 64.Sugumaran R, Burnett J, Blinkmann A (2012) Big 3d spatial data processing using cloud computing environment. In: Proceedings of the 1st ACM SIGSPATIAL international workshop on analytics for big geospatial dataGoogle Scholar
 65.Friedman J, Hastie T, Tibshirani R (2008) Sparse inverse covariance estimation with the graphical lasso. Biostatistics 9(3):432–441CrossRefzbMATHGoogle Scholar
 66.Scheinberg K, Ma S, Goldfarb D (2010) Sparse inverse covariance selection via alternating linearization methods. In: Advances in neural information processing systemsGoogle Scholar
 67.Qiu J, Zhang B (2013) Mammoth data in the cloud: clustering social images. Clouds Grids Big Data 23:231Google Scholar
 68.Hoi SC et al (2012) Online feature selection for mining big data. In: Proceedings of the 1st international workshop on big data, streams and heterogeneous source mining: algorithms, systems, programming models and applicationsGoogle Scholar
 69.Hartigan JA, Wong MA (1979) Algorithm AS 136: a kmeans clustering algorithm. In: Applied statistics, pp 100–108Google Scholar
 70.Wold S, Esbensen K, Geladi P (1987) Principal component analysis. Chemometr Intell Lab Syst 2(1):37–52CrossRefGoogle Scholar
 71.Azar AT, Hassanien AE (2014) Dimensionality reduction of medical big data using neuralfuzzy classifier. Soft Comput 19(4):1115–1127CrossRefGoogle Scholar
 72.Cichocki A (2014) Era of big data processing: a new approach via tensor networks and tensor decompositions. arXiv preprint arXiv:1403.2048
 73.Dalessandro B (2013) Bring the noise: embracing randomness is the key to scaling up machine learning algorithms. Big Data 1(2):110–112CrossRefGoogle Scholar
 74.Zeng XQ, Li GZ (2014) Incremental partial least squares analysis of big streaming data. Pattern Recogn 47(11):3726–3735CrossRefGoogle Scholar
 75.Ruhe A (1984) Rational Krylov sequence methods for eigenvalue computation. Linear Algebra Appl 58:391–405MathSciNetCrossRefzbMATHGoogle Scholar
 76.Tannahill BK, Jamshidi M (2014) System of systems and big data analytics–Bridging the gap. Comput Electr Eng 40(1):2–15CrossRefGoogle Scholar
 77.Liu Q et al (2014) Mining the big data: the critical feature dimension problem. In: 2014 IIAI 3rd international conference on advanced applied informatics (IIAIAAI)Google Scholar
 78.Jiang P et al (2014) An intelligent information forwarder for healthcare big data systems with distributed wearable sensors. IEEE Syst J PP(99):1–9Google Scholar
 79.Leung CKS, MacKinnon RK, Jiang F (2014) Reducing the search space for big data mining for interesting patterns from uncertain data. In: 2014 IEEE international congress on big data (BigData congress)Google Scholar
 80.Stateczny A, WlodarczykSielicka M (2014) Selforganizing artificial neural networks into hydrographic big data reduction process. In: Rough sets and intelligent systems paradigms. Springer, pp 335–342Google Scholar
 81.Hinton GE, Osindero S, Teh YW (2006) A fast learning algorithm for deep belief nets. Neural Comput 18(7):1527–1554MathSciNetCrossRefzbMATHGoogle Scholar
 82.LeCun Y et al (1998) Gradientbased learning applied to document recognition. Proc IEEE 86(11):2278–2324CrossRefGoogle Scholar
 83.Kavukcuoglu K et al (2009) Learning invariant features through topographic filter maps. In: 2009 IEEE conference on computer vision and pattern recognition, CVPR 2009Google Scholar
 84.Dean J et al (2012) Large scale distributed deep networks. In: Advances in neural information processing systemsGoogle Scholar
 85.Martens J (2010) Deep learning via Hessianfree optimization. In: Proceedings of the 27th international conference on machine learning (ICML10), June 21–24, Haifa, IsraelGoogle Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.