Abstract
Universities and colleges aim to provide students with a solid academic foundation. Quality instruction is one strategy for achieving the highest possible standard in the higher education system. Personalized teaching caters to each student by adapting the learning pace and method to their specific requirements. However, the present state of customized education in higher education resources prevents proper resources from being extracted due to a lack of multi-dimensional association analysis between students, circumstances, and materials. A hybrid personalized teaching system utilizing fuzzy association rules mining is the goal of this research to improve learning in higher education. Effective multi-dimensional association analysis among students, settings, and instructional materials is facilitated by the fuzzy association rules mining-based hybrid personalized teaching system (FARM-HPT). The proposed study conforms to AI standards, is based on fuzzy logic theories, and guarantees precise university-level resource discovery. The study builds on earlier work in data mining by presenting a new, learner-specific recommendation model for personalized teaching that uses FARM to ensure accurate resource recognition and efficient mining of instructional assets at the higher education level. This new approach generates fewer set comparisons and does them faster than the current standard. Focusing on experimental validation, the study shows that the FARM-HPT system can generate individualized lessons while overcoming the constraints of traditional information mining methods. These findings align with AI standards, which shows how important it is to validate new AI approaches using robust empirical evidence. The system ensures effective accuracy on various datasets: LFW (89.76%), JAOLAD (94.43%), OECD (95.43%) and OULAD (97.45%).
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Avoid common mistakes on your manuscript.
1 Introduction
The instructional approach is used in the individual teaching model of higher education based on several different pedagogical philosophies. A customized education program tailored to students’ prior knowledge and preferred learning style [1]. Increased student engagement and motivation, higher student achievement and academic performance, and significant use of time and resources are benefits of personalized instruction. The interactive teaching paradigm emphasizes the instructor as the prominent figure rather than the students [2]. The interactive method of instruction places a greater emphasis on students’ cooperative learning spirit and fosters students’ individual growth. In its simplest form, personalized education occurs when a teacher presents information to a student in a manner that considers their unique background and learning style. The higher education sector is moving toward a more individualized approach to instruction that draws from a wide range of pedagogical theories and practices [3]. Teachers are the dominant figures in the interactive teaching model, which places students at its centre. The interactive method of instruction places more excellent value on students’ individual growth and teamwork in the classroom [4]. Fuzzy Association Rules Mining is one of the artificial intelligence methods that the model incorporates to make higher education-tailored instruction more efficient and accurate.
FARM effectively handles ambiguity and represents newly acquired information [5]. The rate at which educators and their students use personalized teaching resources has increased, and the exponential growth of available resources has made it harder to zero in on the most useful ones. Consequently, the importance of individualized educational materials cannot be overstated [6]. A personalized instruction system is one of the most recent ideas implemented effectively in higher education, emphasizing the individual learner in the methodology. Several service technologies are involved, from the user-specific configuration of learning scenarios to the use of hybrid filtering recommendation technology to the content of educational materials [7]. Regarding domestic educational resources, data mining and learning behaviour analysis are accomplished with the help of a recommendation model and tailored teaching technologies. Interest data are gleaned from user profiles to assess students’ levels of motivation and cognition, facilitate the exchange of feedback on pedagogical materials, and so on [8, 9].
The research adds fuzzy set theory to the conventional association rule mining (ARM) approach, yielding significant methodological improvements. Association rules, reinforcement, and trust are developed for classroom teaching. The generation of association rules relies heavily on these two settings. Rules with a low support value may be generated by coincidence, providing less satisfaction. When both the support and confidence values of an association rule are higher than the minimal support and confidence thresholds, respectively, the rule is considered intriguing. The statistical significance and interest of the association rules produced are guaranteed by the thorough examination, which is typical of the methodological rigour that characterizes artificial intelligence.
Additionally, it broadens the ARM by factoring in students’ requirements and giving importance to resource material and the standard of higher education. According to previous records, the Apriori procedure in FARM allows for the inference of the learner’s desire for individualized instruction [10, 11]. As mentioned earlier, the methods are arduous and time-consuming since they cannot collect the multifarious association rules of data and make several blunders while obtaining the most vital elements regarding instructional materials [12]. This research develops a fuzzy association rule-based framework for the FARM approach to higher education resources. FARM-HPT is a method developed for personalized teaching in higher education numerical data found in several databases effectively. The most crucial information from teaching aids is extracted, and then similar data are clustered using the various values of the feature dataset clustering approach. Due to the large quantity of data of an educational resource set and the lack of performance of conventional data mining techniques, it is necessary to enhance and achieve optimum mining of instructional materials. Construct a college student preference matrix, utilize the matrix deconstruction to develop an exclusive asset suggestion model, and then utilize this model to guide the development of individualized curriculums [13]. HPT recommender system used in the model should also assess a method of instruction suggestion system for learning networks that accounts for students’ preferences and unique educational needs. In the framework of differentiated pedagogy, it is essential to acknowledge the diversity of students’ cultural backgrounds and to provide accommodations for promoting those cultures and equitable consideration of children from various cultural communities. Numerous techniques that utilize data mining, systems for recommendation, and clustering are addressed in current studies across personalized teaching from higher education; however, they still fall short in adapting to learning frameworks and domain expertise, accuracy, and efficiency.
To address this, FARM-HPT offers novelty by fusing personalized recommendation models and fuzzy association rules to provide more effective and specialized teaching procedures [14]. This article presents a novel method for individualized higher education instruction based on fuzzy association rules mining (FARM). Integrating FARM, a computational intelligence technique, with the hybrid personalized teaching system (FARM-HPT) improves the learning environment at the university level. The present educational paradigm has a hole that this system fills by integrating hybrid and instructional instruction with pre-training evaluation. The theory of fuzzy sets and rules for the association are combined in FARM, which has advantages, including enhanced precision and effectiveness in creating personalized pathways to learning for students, emphasizing computational intelligence techniques and their application in diverse educational settings [15]. FARM-HPT draws attention to the lack of multifaceted analysis of associations and stresses the necessity of eschewing one-size-fits-all approaches. The study emphasizes speed, higher efficiency, and accuracy in making suggestions. Obtaining a respectable lower RMSE in personalized learning recommendations shows its applicability and efficiency, showcasing its effectiveness within the context of the computational intelligence landscape.
This paper’s contribution is listed as follows.
-
1.
To enhance the education quality of higher education students using fuzzy association rule mining based hybrid personalized teaching system (FARM-HPT).
-
2.
To extract the education resources for personalize the teaching factors using the interval value attribute dataset.
-
3.
FARM data mining methods, such as partition-based strategy and byte vector representation, are proposed to optimize educational resource mining for the HPT recommendation model due to the enormous quantity of information in the higher education material set of period value characteristics.
-
4.
With a low root-mean-squared error (RMSE) and a data analysis time of 19 ms, the study results show that the suggested model of individualized training using educational resources can compensate for the limitations of conventional data mining techniques.
The following portions of this work are organized. Section 2 summarizes the current status of research and related work on FARM. The inner workings and current curves of FARM and the recommended framework for issue detection in individualized education are discussed in Sect. 3. Section 4 discusses the experimental results. Finally, Sect. 5 brings the paper to a close.
2 Literature Survey
Gao et al. [16] presented an artificial intelligence-based customized recommendation system (AI-CRS) for English teaching materials. First, data about individual users are collected using web mining software and rules and patterns from relevant educational assets are extracted for potential usage as representations of user characteristics. Then, the most similar instructional materials are picked as the recommendation target to execute resource suggestions, and the suggested format is selected. Based on the results of the studies, it seems that this technique improves the precision of information recommendations and makes it possible for educational materials to significantly meet individual users’ needs.
Li et al. [17] proposed a model of public opinion on distance education using fuzzy association rules and cross-dimensional data mining. Using the Yaahp software, researchers may analyze public opinion’s subject, object, and ontology to understand how it relates to online teaching and learning. To understand the interplay between the various components of public opinion data, the results of a weighted analysis are factored into the interpretation of data semantic associations. A cross-dimensional extraction model of the general public in distance learning data is created using multi-dimensional correlation analysis and data pre-treatment and implemented on a cloud computing platform. The simulation findings demonstrate that the model may serve as a reliable source of information to back up efforts to transform education at the grassroots level.
Wang et al. [18] proposed a dynamic, collaborative filtering algorithm-based approach to recommending educational resources for individual students. The changing k-nearest-neighbour and slope algorithms assist in optimizing the cooperative filtering method. However, there is just one strategy for handling sparse data. Its scalability is evaluated by looking at the learning resource data in the network and how selective its neighbours are. To this end, we developed a fuzzy adaptable binary particle swarm optimization algorithm (FABPSOA) that uses evolutionary state judgment to improve individual recommendations for resource deployment by resolving issues like optimal sequence recommendation and bidirectional self-equalization at intermediate stages. Analysis of experimental data shows that the proposed approach enhances recommendation speed and similarity degree.
Komleva et al. [19] proposed a software system that can be used to build adaptive online courses for individual students and integrate them into the university’s existing e-learning infrastructure. Helping students meet their field’s educational and professional criteria, the Digital Tutor platform is equipped with adaptive testing tools that can tailor the electronic course material to each learner’s specific level of knowledge and skill. Making use of cutting-edge digital and intelligent technology is critical to the idea behind revamping the paradigm of online education. With a mechanism for automatically updating educational content and a repository of educational objects that formed the necessary competencies following the Federal State Educational Standard and approved professional standards, the project has competitive advantages in implementing a new education business model.
Fedushko et al. [20] introduced a recommendation system decision-making architecture to aid newcomers in choosing a niche. The system relies heavily on the modelled database to comprehensively explain academic subfields. Exploratory data analysis of the 2021 Ukrainian university admissions drive validated the predictions and revealed new information regarding newcomer specialization selection. As a result, higher college enrolment and professional development rates can be achieved through smart algorithms, user behaviour statistical analysis, and academic and career planning experts. The data will create an intelligent virtual assistant for the entrant.
Dogan et al. [21] proposed association rule mining with fuzzy logic for increasing profits (P-FARM), which considers the company’s bottom line while providing recommendations. P-FARM is a state-of-the-art data mining technique for mining valuable information from clusters of frequently used items to produce association rules. From the practitioners’ perspective, this approach aids businesses in decision-making by giving them access to more lucrative items with fewer restrictions. This research demonstrates that P-FARM is an effective method for increasing revenue in online stores.
Papakostas et al. [22] offer an adaptive training system called PARSAT (Personalization of the learn path in a spatial ability training application for augmented reality utilizing fuzzy weights), which was developed to support three spatial skills-related engineering undergraduate courses over a single semester. PARSAT can modify instruction for students of different cognitive abilities by combining fuzzy weights technologies from a rule-based decision-making module with learning theory from the framework of recorded learning results. Based on the findings, the proposed training approach significantly outperformed the status quo regarding student learning outcomes, which had previously lagged due to a lack of adaptability in domain knowledge and learning theories.
Tran et al. [23] presented a node list, and pre-order size code data structures were used to mine fuzzy association rules. These two formats handle fuzzy data well. Node-list Pre-Order Size Fuzzy Frequent (NPSFF) is a hybrid method. Thus, tree building is accelerated, and fuzzy groups are recognized. Using Affinity Propagation clustering (AP) to evaluate data and calculate the appropriate number of clusters improved NPSFF performance. Fuzzy splitting converts raw cluster numbers into fuzzy numbers. Compared to FFP Growth and MFFP, NPSFF was faster and used less memory.
Modak et al. [24] introduced fuzzy clustering on the gamma-ray bursts in the final catalogue from the ‘Burst and Transient Source Experiment’ to double-check the relevance of these new groups. Cluster studies are undertaken to uncover homogenous groupings, which leads to discovering a third kind of gamma-ray burst beyond the more common short and long bursts, which may indicate the astrophysical origins of these phenomena. The method uses the robust ‘FANNY algorithm’ statistical machine learning technique, which makes no model assumptions and can reliably disclose the underlying clusters in the various gamma-ray bursts. Recent research suggests that there may be more than three distinct groupings. Careful analysis of individual bursts according to their memberships in the fuzzy clusters validates the three well-established groupings versus the five newly discovered ones.
Asadi et al. [2] introduced fuzzy association rule mining to analyze the interrelationships among their course preferences—clustering groups of students by interests and talents. Clustering and fuzzy association algorithms provide us with credible recommendations and projected ratings. This research uses data from undergraduates at the University of Tehran College of Farabi Management and Accounting. Data covers 2004–2015. First, two sets of students—one with a hidden past and one with an average one—are formed. Students then get their proposed assignments and grades. The discovered parameters make choosing an academic career simpler.
Borlea et al. [25] proposed a Unified Form (UF) clustering method, proposed in this study, combining the Fuzzy C-Means (FCM) and K-Means (KM) methods into one customizable method. UF, which can be FCM or KM, facilitates the software execution of FCM and KM algorithms. Another novelty is this paper’s Partitional Interpretation of the United Form (PIUF) approach. It uses the UF algorithm to solve the difficulties of sequentially processing massive datasets and scalability to handle datasets of any size elegantly. The authors’ distributed platform, BigTim, builds and verifies the UF and PIUF algorithms, which may be applied to any data processing platform. Three performance indicators assess clustering excellence, while two compare the PIUF technique to FCM, KM, and DBSCAN clustering methods.
Cui et al. [26] introduce an intelligent recommendation system-based data mining to enhance the efficacy of data mining. The intelligent recommendation system in this study is built on the mathematical modelling of association rules. Java 2 Platform, Enterprise Edition technology divides the system architecture into the display, business logic, and data layers based on intelligent recommendation system requirements. Fuzzy clustering optimizes. According to research, this system has the fastest response time (0.2 s) in the Q1–Q5 subsets, the lowest mean absolute error (MAE; maintained at around 0.73), and the highest accuracy (before the proposed threshold reaches 0.5). After going online, click-through rates increased by 11.04% and conversion rates by 9.35%. 43% of 1216 customers were pleased, and 4.9% were pleased with 1. The algorithm converges quickly and tailors product suggestions to individual users, enhancing click-through, conversion, and user satisfaction.
Niknam et al. [27] introduced a learning path recommendation (LPR) system that would categorize students into groups with similar levels of knowledge and then propose courses of study to them. Clustering is done using the Fuzzy C-Mean (FCM) approach, which may offer multiple learning routes to students outside a cluster. The ACO route finder incorporates learners’ progress and finds an optimal learning trajectory using bio-inspired ant colony optimization (ACO). Finally, provide actual students with a database course to test the LPR system for education. The LPR group outperformed the control group in course performance and comprehension. These findings imply the LPR system significantly impacts student outcomes.
Rohidin et al. [28] proposed Class-Based Fuzzy Soft Associative (CBFSA) for large amounts of information in the text categorization issue may lead to them. Marketing, security, biology, and more utilize it. However, processing speed and accuracy are significant challenges. This model combines association rules with a fuzzy soft set. Fuzzy Soft Set Association Rules Mining and FP-Soft Set Fuzzy Reasoning Set filters. On the 20 Newsgroups dataset of 20 class documents, CBFSA outperformed SCC, FSSC, and HFC in accuracy testing. CBA and other associative classifiers are less accurate and efficient than CBFSA.
Tutsoy et al. [29] introduced self-supervised machine learning techniques to identify the health risk classification. During this process, a blood test is conducted, and the data is explored using a feature selection approach, which selects the most relevant features. The selected features are processed using a deep learning approach that recognizes health risks with a maximum recognition rate. The suggested model was tested on multiple learner cohorts and information sets before tailoring instruction to each individual’s preferred teaching method, interest profile, and innate abilities. When comparing the performance of the suggested PIUF, collaborative filtering algorithm-based approach to recommending educational resources, LPR CBFSA and P-FARM to that of learners in the fuzzy association rules and cross-dimensional data mining, it was found that the fuzzy significantly improved suggestion performance. The present paper suggests a Fuzzy Association Rules Mining-based Hybrid Personalized Teaching System (FARM-HPT) to fill the gap left by current approaches to implementing incorporated typical instruction by optimizing learning creation handles for the execution of distinguished pedagogical programs. Using this combined process of fuzzy decision-making, provide specific suggestions for implementing established, interconnected lesson plans, which are the main tool for building adaptive, different educational programmes adapted to each learner.
Lin et al. [30] suggested the ELECTRE II method to deal with probabilistic linguistic term sets and their application to edge computing. The first step is to create a new distance metric for PLTSs and an entropy metric to quantify the degree of uncertainty in PLTSs. Two PLTSs may be compared using a new approach that relies on the score value and entropy. After that, the author provides a technique for experts to determine their weight using entropy theory and a way for criteria to determine their weight using multiple correlation coefficients. A novel probabilistic linguistic ELECTRE II approach is proposed to address the edge node selection issue. A comparison with earlier approaches shows the method’s superiority.
Zhang et al. [31] proposed the Cache Reallocation-Based Page-Level Flash Translation Layer for Smartphones. Based on an analysis of the I/O request life cycle for smartphone applications, CRFTL uses several mapping cache structures to store the mapping entries of I/O requests with varying life cycles. CRFTL can understand the internal workings of NAND flash memory and the patterns of I/O requests; it then uses a combination of heuristics and reinforcement learning techniques to adaptively modify the allocation of cache space in the mapping cache structures. The suggested CRFTL outperforms state-of-the-art FTLs, according to the experimental data.
Lin et al. [32] recommended the linguistic q-rung orthopair fuzzy sets and their interactional partitioned Heronian mean aggregation operators. With the partitioned geometric Heronian mean (PGHM) operator, decision-making issues with interrelated clustered features and non-interrelated attributes may be efficiently solved. These new operational rules and the PGHM operator form the basis of the proposed linguistic q-rung orthopair fuzzy interactional PGHM (LqROFIPGHM) and linguistic q-rung orthopair fuzzy interactional weighted PGHM (LqROFIWPGHM) operators, which are then explored in terms of their characteristics. An effective multi-attribute group decision-making model handles linguistic q-rung orthopair fuzzy data using the LqROFIWPGHM operator at its core. The interactional operational rules and the LqROFIWPGHM operator are compared and contrasted with the help of many instances towards the end.
Lin et al. [33] discussed the Pythagorean fuzzy sets (PFSs) for medical diagnosis and cluster analysis. To address the shortcomings of the current PFS correlation coefficients, we propose new directional correlation coefficients that assess the relationship between two PFSs by taking into account the degree of membership, the degree of non-membership, the strength of commitment, and the direction of commitment. Then, two real-world examples illustrate how the suggested directional correlation coefficient may be used in illness detection and how the proposed weighted directional correlation coefficient can be used in cluster analysis. The last step is to compare them to the earlier PFS correlation coefficients.
Lin et al. [34] deliberated the Picture fuzzy interactional partitioned Heronian mean (PFIPHM) aggregation operators for multiple attribute decision-making (MADM). To calculate picture fuzzy numbers (PFNs), which may represent the interplay between agreement degrees (AD), neutrality degrees (ND), and opposition degrees (OD) in two PFNs, this research suggests using interactional operational laws (IOLs). To illustrate the decision-making process of the proposed MADM technique and to compare the suggested operators to the current aggregation operators created for PFNs, a study case involving the service quality rating of nursing facilities is offered.
All of the efforts mentioned above are directed at enhancing the quality of education by enhancing tailored teaching techniques and technological capacities. In addition, some models enhance individualization by providing more opportunities for interactive study, which has been shown to boost motivation. Successful models are almost always founded, at least in part, on fuzzy association mining techniques.
3 Fuzzy Association Rule Mining for Personalized Teaching from Higher Education
3.1 Descriptive Analysis and Model Development
Providing hybrid personalized education requires considerable preparation and focus on individual student needs. The method of instruction must be carefully aligned with necessary educational outcomes, informed in progress evaluation, responsive to student readiness, interest, and learning profile, and flexible groupings based on the deliberately balanced person. More specific than differentiated instruction is individualized or personalized instruction, defined as an institution’s concerted effort to structure its teaching recommendation environment to accommodate each student’s unique characteristics and demand for adaptable teaching strategies. As the country has focused more on education in recent years, informatization of managing learners has become increasingly crucial. With the widespread adoption of the Internet plus education, a wealth of new digital educational tools has become available. While there is no shortage of resources available to online students in higher education, less consideration is given to the diversity and relevance of these materials, making it more challenging for students to find helpful information. Learners need help gauging knowledge and progress, making it simple to become disoriented to make it forward. The ideal online learning system would use an intelligent algorithm to implement learning materials through individualized instruction and a recommendation system and provide tailored recommendations for resources based on student attributes. Personalized teaching–learning resource recommendation systems must improve speed and match because learner attributes are volatile and hard to define, and sophisticated materials are available. Figure 1 illustrates the structure of the FARM-HPT system. This essay’s topic is rebuilding personalized education with the help of fuzzy association rule mining. The research uses a data-mining-based strategy provided for a fuzzy association, demonstrating its application to extracting knowledge about instructional materials. Teachers in special education classrooms are ultimately responsible for the program’s success or failure. The hybrid customized recommendation system allows students to use the intelligent instructor’s data. The FARM extracts data for the recommendation system by clustering and then distributes those resources. Their dedication to curriculum creation is crucial for ensuring the success of special education programs.
Currently, multifaceted collaboration evaluation of learners, circumstances, and assets is impossible, and accurate resources are not extracted for learners using the personalized teaching method of higher education resources. In this research, FARM develops an individualized strategy for recommending reading materials for college students. Fuzzy association rules are simplified, the relevant data are extracted from extensive educational resources, and data of the same sort are clustered using the Apriori algorithm based on the regular item set method of fuzzy mining. This study uses a novel, personalized suggestion methodology for educational content based on fuzzy association rules to guarantee accurate extraction of academic materials and a high degree of learner-resource compatibility. Data collection involves gathering detailed information on student performance, preferences, and behaviours. Data analysis includes pre-processing, fuzzification, rule generation, and defuzzification to extract meaningful patterns. The learning process encompasses continually using these collected insights to create and adapt personalized learning paths for higher education.
The input of the proposed model should match student requirements and resource attributes for the best results, which is fundamental to personalized teaching resource suggestions. Individualized recommendations are the focus of the current paradigm for suggesting educational materials whether or not the learning resources’ difficulty level corresponds to the learners’ ability level.
This paper provided an HPT Recommender System for recommending educational resources in a Personalized Learning Environment.
-
1.
This HPT Teaching object recommender system makes recommendations based on student profiles. An HPT recommender system is a collection of programs and procedures designed to help people find resources that interest them.
-
2.
In the HPT, the instructor considers a student’s needs, the program in which the student is presently enrolled, and the significance and excellence of the instructional resources. In addition, the student’s registered courses and degree of self-assurance in those subjects are included in the profile.
-
3.
Research uses collaborative filtering methods to suggest user interactions or intriguing shared resources. The collaborative HPT recommender model aims to make educated predictions about user preferences and resources.
-
4.
Finally, evaluations of objects from implicit and explicit learning should be included to improve the suggestions. Virtual learning environments have been the subject of several studies in recent years, but their ability to cater to each student’s needs remains constrained. Instead, using such resources improves education by making an HPT recommender system that uses machine learning to provide individualized learning paths for each student.
The extent to which the material types preferred by the students and the learning ideas covered by the available resources for higher education are compatible. This aspect is consistent with students’ choices for distance education. Fewer learners will have significantly different resource sequences. The group recommendation phase progresses into the personalized recommendation phase, which may be characterized as the process by which the suggested learning resource sequence becomes more following the individual’s specific demands. The instructor’s input, which the students must accept, comprises two parts: characteristics of the learners and elements of the resources. These parts work together to form an equilibrium with capabilities based on personalized teaching suggestions and the accuracy of the instructor’s recommendations of personalized instructional materials.
3.2 Fuzzy Association Rule Mining for Personalized Teaching
A practical and quick approach for mining fuzzy association rules in personalized learning, even for massive datasets. Association rule mining converts numerical features into fuzzy qualities without compromising the accuracy of the instructional information they provide. Previous work in association rule mining has given techniques for mining quantitative rules of the association. However, the sharp boundary issue arises when data is divided along an attribute’s range of values. Fuzzy logic has been used in association rule mining to solve this issue. When tested on a large, conventional, real-world dataset, the suggested method is faster than fuzzy Apriori on various standard and nonstandard mining tasks.
Generating fuzzy association rules using a suitable fuzzy ARM method is complex. At first, transform the crisp dataset, which only has numeric and binary properties, into a fuzzy dataset with binary attributes. Extensive pre-processing utilizing suitable methods, such as the one outlined, is required. Among the steps in such pre-processing is transforming numerical characteristics into fuzzy binary attributes. Second, while crisp ARM algorithms can determine whether a given item set occurs in a dataset by checking whether or not it appears in a given transaction, fuzzified ARM algorithms account for the item set’s fuzzy membership. Although standard ARM algorithms can generate crisp association rules from the same dataset, employing a FARM method to construct fuzzy association rules is far more extensive and sophisticated.
The hybrid Apriori algorithm has an unusual mix of characteristics, including couple-phased multiple-data processing, octet-vector encoding, and rapid compression, significantly improving speed for learners in higher education. In addition, the second phase is quite distinct from the first (processing each item in the set individually instead of simultaneously at each k-level, as with most couple-phased ARM algorithms) and is much quicker. An efficient preparation technique for transitioning from a clean to a fuzzy dataset is included in the system. The framework of the FARM approach proposed to facilitate personalized higher education instruction is discussed. There are four modules: Fuzzy pre-processing Methodology, data mining discovery module, FARM evaluation module, and decision support module.
3.2.1 Fuzzy Pre-processing Methodology
In fuzzy pre-processing, before the implementation of FARM, a hard dataset D is transformed into a fuzzy dataset E, using a data-driven pre-processing strategy that uses automation to generate fuzzy divisions for numerical features.
-
1.
First, converting complex data into a membership function may be used to create a fuzzy model. For example, crisp input x ∈ U is mapped to a fuzzy set A ∈ U by fuzzification.
-
2.
Step two involves analyzing the crisp input x to establish the degree to which it fits into each of the relevant fuzzy sets. Finally, rule evaluation refers to using rules to ascertain what control actions should be taken in response to inputs.
-
3.
Third, defuzzification is the last step after rule assessment.
Another option is to utilize fuzzy clustering to automate the generation of fuzzy partitions. Partitions can be accomplished with minimum human effort, even for massive datasets. The recommendation is to build piecewise linear fuzzy partitions with the help of data-driven fuzzy c-means clustering. Fuzzy partitions are automatically produced for numerical properties. Most real-world datasets include numerical data, which may be represented as Gaussian-like fuzzy sets, with each data point potentially belonging to fuzzy sets. A data point’s reliability and precision of the fuzzy association rules produced from it might be affected if it is a member of more than two fuzzy sets simultaneously. Most fuzzy sets of characteristics in real-world datasets are Gaussian in shape rather than a perfect triangle. Setting the fuzziness parameter m to a suitable value governs fuzzy sets’ degree of fuzziness and Gaussian distribution. Figure 2 illustrates the first two modules of the FARM technique. Teaching materials and knowledge, the role of the lecturer, and student participation and assessment for learning are all preserved in a data warehouse. FARM parameters are determined and extracted from the data warehouse in this module. Before entering parameters into the mining algorithm of the Data mining Discovery process Module, imprecise characteristics of parameters must be defined. This includes ambiguous terms used to describe the quantitative values of parameters and member functions that account for the fuzziness of parameters—the methodological framework for fuzzy pre-processing and fuzzy measurement used in the practical application of FARM. There are two stages to this pre-processing method: fuzzy-partitioning-of-numerical-attributes creation and typical fuzzy data representation techniques used to transform an exact dataset into a fuzzy one, which is represented in Fig. 2.
3.2.2 Data Mining Discovery Module
In the first stage, the FCM (Fuzzy C-means) approach is used to acquire a cluster on each quantitative variable. Inside that cluster, each numeric attribute value is individually recognizable by its membership functionality in these fuzz partitions. A proper choice of k allows the clusters to be labelled according to the characteristic. Fuzzy C- indicates that in pattern recognition, a single data point may belong to several clusters. A finite collection of points is partitioned using predetermined criteria into C fuzzy groups. Therefore, the centres of clusters may not be the most central spots. The FCM method is based on minimizing the objective function described below:
Fuzzy partitions are automatically generated for numerical properties using a data-driven pre-processing technique. It aids in fuzzy dataset partitioning, where each data point is partially (in the range [0, 1]) associated with each cluster. It means there is room for more than one cluster for every piece of data. As seen in Eq. 1, the algorithm seeks to minimize the partitioning objective function.
where \({\mu }_{ij}\) represents the value of membership of xi on cluster j, xi represents an ith degree of the information being measured, and dj represents the cluster’s centre in that dimension. M is the real integer (m > 1) that may be used for the fuzziness variable. Each instance is wholly connected with one cluster and partially correlated with all of the clustered centroids, thanks to fuzzy c-means (FCM), the most effective fuzzy clustering algorithm currently available. As m approaches infinity, the solution’s fuzziness approaches 1. There is a growing resemblance between the result and the grouping achieved by binary k-means. An appropriate value for m is 2.0.
This method analyzes the gap or overlap between several input data sets. Data points are grouped into clusters based on proximity to their cluster centres. For example, using the FCM clustering method, a dataset may be divided into a maximum of n clusters, each of which can be linked to every other cluster in the dataset. If the cluster’s centre is located distant from the data point, then the connection between the two will be low; otherwise, the degree of connection will be high.
Second, if there is a quantitative attribute, each sharper record in dataset D is converted into numerous fuzz entries according to the number of fuzzy divisions established for the attribute. Fuzzy association records might be generated at a combinatorial exponential rate.
To address this issue, the membership function is set at a lower threshold (K means 0.1) to limit the creation of unnecessary fuzzy records of educational data. If an attribute is binary, attach a membership function of µ = 1 to the end of each record to signify that the value is a complete member of the binary attribute.
3.2.3 Fuzzy Association Rule Mining Evaluation Module
Figure 3 illustrates the function of the FARM method for the hybrid personalized teaching on data FARM and decision assistance module. Rule assessment is necessary due to the iterative nature of defining threshold values. Rules might be developed, some of which could be simple or nonsensical, depending on the specification of threshold values. The large item sets are identified in Apriori’s initial phase by counting occurrences of items. Both of these steps must be completed in any subsequent pass x. The candidate itemsets for the xth pass are first generated using the big itemsets discovered in the (x − 1)th pass. The next step is to check the database and tally the backing for potential item sets. Each record is picked, and for each pass x, the supports for the candidate item sets that appear in that record are boosted by one. In Apriori, numbers are tallied one record at a time. The FARM process converts the original dataset into a longer one with attribute scores in the range [0, 1] so the learners utilize more resources on personalized teaching. Several fuzzy partitions are built on the attribute’s domain for each quantitative attribute. The study’s rules will have the resource characteristics as their antecedents and the instruction as their consequents. Then, the sample rules are listed as follows.
If the student has limited prior knowledge, a visual learning style, and the subject is mathematics, then suggest example-rich introductory video tutorials.
If the student has extensive prior knowledge, an auditory learning style, and the subject is literature, then recordings and audio lectures should be suggested.
If the learning environment is online and the format of the resources is interactive, then interactive exercises and simulations should be suggested.
Any rules that might break this framework are discarded. The system generates fuzzy association rules in two stages using a partition-based strategy. The information set has p conceptually separate horizontal divisions (x1, x2,…, xp). Each partition is as big as can be stored in a computer database. Therefore, define fuzzy association rule support using fuzzy partitions P and Q at transaction x, and a threshold-norm is given in Eqs. 2 and 3
New metrics of crisp-like support (CS) and instructional confidence (IC) in terms of threshold norms are required for processing the enlarged fuzzy rule dataset using Eqs. 4 and 5
P and Q are the set of partitions in a transaction x, and d is the number of partitions.
The threshold value for each group of items is stored in the cell of the byte vector whose index is equal to the threshold value. Therefore, the norms for the ith threshold are stored in the ith cell of the byte vector. As a result, the byte vector will initially have a value of 0 in each column. There is an ongoing reorganization since threshold lists may be extensive, especially with byte-vector encoding. Through the use of a suitable compression strategy that involved fast inflation and deflation rates, we were able to solve the issue. While other ARM algorithms, such as VIPE, have experimented with different compression techniques, none are successful or fast in every imaginable operation setting. The compression method significantly reduced the number of partitions needed to analyze each dataset while achieving significant speedup for various thresholds.
In addition, the parameter update rule is also given to improve the decision-making process and higher education recommendations.
-
If the resource recommendations generated by the system consistently fail to accurately represent a specific student characteristic (e.g., prior knowledge level), the membership functions for that characteristic should be modified following the misclassified instances.
-
If user feedback suggests a consistent misjudgment of the difficulty levels associated with recommended resources, the membership functions for those resources should be modified to reflect the feedback.
3.2.4 Decision Assistance Module
The decision rules learned by FARM are kept in a rule-based information library in the Decision Assistance Module. The guidelines may help instructors get the insight needed to create valuable new teaching resources. By feeding the algorithm data about both the teacher and the educational materials, the algorithm can uncover previously unknown associations between the two sets of data. Learners may obtain some insight into the elements of higher education that can be improved by using such information. The knowledge is expressed in a simple form since the parameters in the rules are defined using fuzzy language concepts.
3.3 Hybrid Personalized Recommendation for Higher Education
Figure 4 represents the proposed FARM-HPT’s hybrid personalized teaching recommendation flow chart. The suggestion module is the backbone of the individualized classroom resource system and is significantly responsible for the success or failure of the whole system. Learning Resources and Learner-List are two inputs of the proposed HPT recommender model. The algorithm generates a higher-quality suggestion list for the students to peruse. Figure 4 shows a high-level flowchart of our proposed HPT recommender. The outline of the suggested model may be seen in Algorithm 1. The inputs to this model are a list of learners and a set of instructional materials. The Collaborative filtering recommendation method must be selected and utilized for learner-based applications to improve the quality of suggestions produced by the HPT recommendation module.
The personalized data can be utilized through the HPT recommendation model, as shown in Fig. 4. There are several ways in which the HPT recommendation system is tailored to the specific needs of particular users in terms of personalized teaching for enhancing knowledge. For example, utility and similarity functions for the case and constraint-based recommender systems may be learned per user. In addition, it may be utilized as input from end users to prioritize specific criteria. The CF algorithm for recommending models has received the most significant attention and is widely used. It is very effective and is based on the needs of the community. Learning Resources are the information or resources broken down into smaller chunks, each consisting of discrete lessons developed by collaborative filtering. Typical lesson structures include introductory and introductory overview materials, application materials, tutorials, assessments, and exercises. In this case, Learner-List is a database that stores teacher and student data. The instructor creates various lesson parts and uses the proper authoring software. Learners may vary in needs, interests, and preferred approaches to education. Characteristics of learning styles are determined by assessing individual preferences along several aspects.
Algorithm 1: Hybrid personalized teaching recommender algorithm
-
Input: Instructor Materials, Student Information List.
-
Output: The result is a profile of the learner’s preferred HPT approach.
-
1: Apply collaborative filtering resources on the instructor learning material and then index it.
-
2: Split up the necessary modules and lessons from the available learning resources. Set definition Items-to-Be-Learned = [Item1, Item2,…, Item n] To define the order of the sets of items, write (Item1, Item2,…, Item n).
3: Provide the different classes’ and modules’ respective examples, quizzes, and exercises.
-
4: Outline the parts needed to supplement existing educational materials
-
5: Determine the various learning approaches depending on student profiles.
-
6: Examine learner preferences across several dimensions
-
7: Process the data and determine how the data will be perceived,
-
9: Determine how various students will receive and understand the data.
-
10: Specify the various student evaluation scales.
-
11: Publish the features of each learning type
The recommender model’s mean square error is calculated by inserting an exponential random integer with variance and mean into the matrix. The matrix decomposition-based technique is used for a revised recommendation of higher education resources. Then, the square error is updated until an appropriate number of iterations are reached. Here is the broken-down method. Input data consists of a matrix of scores and several attributes of college students. Consider the following student interest matrix for a given university is represented in Eq. 6
Gi,j stands for the variables present in the ith and jth rows of the matrix, while m and n stand for the number of students and the total number of items that may be suggested. Consider a square matrix approximation Eq. 7
\(\delta \) is an optimistic actual number bigger than 0 and less than 1. Finding a low-rank matrix that approximates well enough of matrix \(\vartheta \), where 0 and 1 are disregarded. The norm loss function of minimizing weight W may be used to determine which value of N comes closest to the supplied matrix Y in Eq. 8
\({G}_{i,j}-{\vartheta }_{i,j}\) represents the average error of an input information point to the loss function in the N-by-N low-rank matrix. Gij = 1 for all positive cases, while Gij = 0 when all blanks are unfavourable. Weights are assigned to effective instances due to their substantial certainty. Although it’s not always the case that negative examples are drawn from mixed data, the data often gets a negligible weight of about [0, 1].
An individual recommendation for the approximating matrices for square errors is generated, and the time overhead of the personalized suggestion method for university assets is studied. The number of attributes is assumed to be Z, the total number of duplicates to be Z′, the number of students to be N, and the number of resources to be N. Matrix multiplication is used to determine the time complexity of the method for providing personalized curriculum recommendations. Instead of speeding up the process, collecting resources for higher education slows it down. Equation 9 is the mathematical representation of the model for tailored recommendation suggestions.
The suggested approach combines the time-intensive collecting of academic materials with the time-intensive process of suggesting higher education resources to maximize harvesting educational assets. To further tailor instruction to each individual, it applies fuzzy rules and the Hybrid Apriori algorithm to my extensive database of fuzzy association rules.
4 Results and Discussion
The proposed model utilizes the Open University Learning Analytics Dataset (OULAD). https://www.kaggle.com/datasets/anlgrbz/student-demographics-online-education-dataoulad [35]. Open University students who are not physically present at the institution may use the Access course materials, have virtual discussions, hand in and get feedback on assignments, and more with the help of a VLE. There are a total of seven classes included. Location, age, disability, education, gender, and sexual orientation are just a few of the student data categories collected. The outcomes of the students’ virtual learning environment (VLE) activities and assessments are also provided. Education as a whole and online learning, in particular, have a lot of space for development regarding quality and efficiency. It would be interesting to discover more about how people’s learning methods vary. There have been 23,088 views and 2472 downloads. https://www.kaggle.com/datasets/atulanandjha/lfwpeople?resource=download [36]. The dataset includes almost 13,000 web-collected facial photos. The names of the people depicted next to their respective faces are listed below. We have more than one photo of 1680 of the persons pictured. These faces only have one requirement: they had to be picked up by the Viola-Jones algorithm. https://www.kaggle.com/datasets/junyiacademy/learning-activity-public-dataset-by-junyi-academy [37]. Online education has become increasingly vital and popular as the digital age progresses. The present COVID-19 situation demonstrates the importance of online learning and how it can serve as a viable medium for equal-quality education while campuses are closed. The Junyi Academy Foundation is delighted to support our academic communities during this pandemic. This Taiwanese nonprofit was founded to provide all children access to high-quality education through technology. More than 72,000 individuals used our platform to record their exercise attempts over a year (from August 2018 to July 2019), and we’ve made that data publicly available. We believe our dataset has the potential to further studies aimed at developing more effective and tailored approaches to online education, and we hope that this will lead to increased engagement from experts from various fields. https://www.oecd.org/education/database.htm [38]. The OECD’s growing statistics database includes learning at a Glance statistics. Country, year, and topic filters are available for locating values. A series of trend markers are also accessible to supplement these data sets. Due to discussions with OECD countries, some of the definitions and scope of this online database have changed. The raw data from which the learning at a glance indicators were calculated are also included in this repository. The information in the database is based on national administrative sources submitted by education ministries or national statistical agencies and organized using international definitions, categories, and standards. The data collected each year focuses on the social and economic effects associated with schooling, the outputs produced by educational institutions, the financial and human capital involved in education, and the structural aspects of education systems. The AI-CRS (Artificial Intelligence-based Customized Recommendation System), FABPSOA (Fuzzy Adaptive Binary Particle Swarm Optimization Algorithm), and P-FARM (Profit-supported Association Rule Mining with Fuzzy Theory) are all used as benchmarks to ensure that the experiment is as thorough and accurate as possible. The simulation platform is the Windows 13 OS, while the development platform is MATLAB 2022. The primary frequency is 3.20 GHz, and there are 6 GB of RAM and an Intel Core Processor i6-4570 in the hardware setup. Experimental sample data is used to calculate indicators like the Fuzzy C-means Clustering error rate in higher education, the efficiency of FARM, the time required to implement a personalized teaching recommendation model, and the FARM-HPT’s accuracy and RMSE in extracting essential information from educational resources.
4.1 Fuzzy C-Means Clustering Time Complexity in Higher Education
The temporal complexity of fuzzy clustering (fanny) and fuzzy c-means was compared in Fig. 5 using both a constant and a dynamic number of clusters. The default parameters for running the fuzzy c-means R program included a maximum of 100 iterations, a Euclidean distance measure, k = 3 for the initial set of centroids, and a data frame as input. On the other hand, m, the fuzziness parameter, depends on the situation and the thing being fuzzy. The value of m = 2 is used in several existing publications, allowing for the derivation of ij. Therefore, we set out to find, starting with m = 1, the optimal value of m for the cites core subset dataset. However, with these constraints, FCM could not obtain a clustering solution. When k was set to 3, FCM did not award a score value to any cluster, and membership values were shallow when m was raised from 1.1 to 1.5. The membership values obtained by fuzzy c-means for a particular dataset have a minimum value of 1/K and a maximum value of infinity as m grows. Remember that the distances between score items and the cluster centroids influence the membership values µij—the estimated physical proximity of cluster centres to the objects in complex datasets. As a result, there is a greater possibility that, as m changes, so do the membership values and CV of the FCM describing the collection of distances between objects. We utilized the FCM technique to learn how the membership values change as m goes from 1 to 2. The software iteratively adjusts the cluster centres and the membership values for each data point, moving the cluster centres to significant places in the dataset. Throughout each cycle, the target function is minimized; in this case, the distance between any two data points and the cluster centre. For large input sizes N, the time complexity is expressed as O(log N × K), where K is the time needed to execute the statement once.
4.2 Performance Efficiency of FARM
The efficiency graph of the suggested FARM model is shown in Fig. 6. An innovative alternative to the standard fuzzy Apriori technique for fuzzy ARM, suitable for enormous datasets. The tests show that the hybrid Apriori method is much quicker than the classic version. Novel characteristics, such as two-phased threshold processing employing divisions and item sets expressed as byte vectors, have allowed this significant speedup. The directory name, category, and processing duration of the interval characteristic dataset can be obtained once the datasets for educational institutions have been imported, allowing for an analysis of the data concept. Calculating Mining Efficiency Performance Data mining success is quantified by reaching its optimal run rate, and this is measured by the formula: Performance = (Total Count/Flow Time)/Optimal Run Rate. After the information is mined, the logical address changes, the independently mineable target is determined, and the information is sent. The suggested rule mining is the most efficient alternative to standard data mining techniques. The proposed model is compared with the AI-CRS, FABPSOA, and P-FARM, on which the proposed FARM-HPT has a higher performance rate.
4.3 Hybrid Personalized Teaching Recommendation Model Accuracy
In Fig. 7, the suggested technique, the accuracy of the resource suggestion method that incorporates the learner model, compared to others regarding the time it takes to complete a specific task. The suggested technique always completes the prerequisite suggestion for the resource in less when data is scheduled for repeated execution; its frequency of execution is proportional to N times, the time it takes to perform the function once—the suggested approach for recommending educational resources, depending on the specifics of the learner model being used. In addition, the proposed technique preserves fewer association rules under various minimal supports, which helps to keep the complexity of teaching recommendations to a minimum. It has been demonstrated experimentally that the proposed method can easily and adaptably modify the minimal support in each dataset to varied situations, ultimately deciding on the picked set and the frequent set. The experimental findings demonstrate the suggested method’s fast convergence and balanced ultimate running time, enhancing the recommendation model’s accuracy. The proposed model is compared with the AI-CRS, FABPSOA, and P-FARM, on which the proposed FARM-HPT has a higher accuracy rate.
4.4 FARM-HPT’s Accuracy
The accuracy of the suggested FARM-HPT is shown graphically in Fig. 8. The effectiveness of feature combinations for AI-CRS, FABPSOA, and P-FARM approaches is investigated separately based on outcomes from individualized education. The FARM has a more excellent resolution, and the testing will be more accurate. The suggested model’s superior accuracy results from a novel fuzzy ARM method developed for rapid and efficient performance on massive datasets. Such data sets are too large for algorithms that operate entirely in memory. Additionally, to improve the standard of higher education, the proposed model uses several modules of fuzzy rule mining to speed up the previously time-consuming process of making decisions based on experience when recommending resources.
4.5 RMSE
Figure 9 shows that the suggested technique results in a smaller root mean square error when clustering educational resources and that data mining takes less than 50 ms. As the number of neighbours in a cluster grows, it tends to stabilize at a constant value after first decreasing. By completing the users’ scoring vectors, the enhanced hybrid similarity algorithm mitigates the sparsity of scoring information and boosts the system’s reliability. This strategy usually produces worse outcomes for educational resource clustering than the learner-model-integrated resource-recommendation method. The average discrepancy between the recommended and preferred values was determined using the root mean squared error (RMSE) algorithm.
Tij is the goal value for a basic hypothesis j, and xij is the recommended value i for that hypothesis. A personalized teaching recommendation model must have minimal values to be effective. In the ideal situation, there should be an absolute connection between the recommendations and the users’ preferences.
The existing methods and the proposed model are compared across datasets in Table 1. The table shows that the proposed model has a greater efficiency of about 97.45%, which supports the hypothesis that it performs significantly. OULAD’s superior performance is demonstrated across various datasets and model types.
5 Conclusion
As part of the flexible paradigm proposed for integrating hybrid and teaching instruction in higher education, pre-training evaluation is included in creating an individualized learning path. Using data from the field of education, the study’s authors developed a fuzzy association rules mining-based hybrid personalized teaching system (FARM-HPT). Using FARM, a new recommendation model for students is introduced, emphasizing computational intelligence approaches to improve individualized learning tactics in higher education. Compared to the current method of doing things, this new method generates fewer set comparisons and does them faster. Test findings show that the tailored recommendation model of instructional resources overcomes the constraints of conventional data mining methods, obtaining an acceptable RMSE and a 19-ms data processing time. Complete automation of the recommendation system’s modelling system, with improved precision and efficiency, is one of the proposed areas of further study. In addition, it considers the system’s development by automatically optimizing surveys using intelligent categorization techniques in areas of interest. Finally, the next crucial stage in the development of the methods is the transition to an online fuzzy inference system in teaching in the future. Future research can optimize the recommendation methods within FARM-HPT and investigate cutting-edge hybrid strategies that incorporate collaboration and content-based filtering approaches. The FARM-HPT model’s adaptability can be improved by broadening its scope to include a wider range of educational resources, such as interactive media, virtual reality, and engaging material. A promising direction for advancement is the creation of adaptable learning pathways that consider unique learning styles, tempos, and objectives. Students’ comprehension and retention of information are much improved, and their test scores and general academic performance are boosted when educational material is customized to individual learning styles. Understanding complicated learning patterns and developing focused interventions that provide students with proactive assistance and personalized feedback are both made easier with the support of fuzzy association rules. However, there are certain limitations to the proposed approach. These include the difficulty in implementing fuzzy logic systems, the requirement for large amounts of initial data for accurate rule mining, and the possible difficulty in constantly updating and maintaining the system to accommodate changing educational needs and diverse student populations.
Additionally, adding metrics for student participation and sentiment analysis may offer more information about the suggested resources’ effectiveness. The future study of longitudinal learning analytics is to track and evaluate development over time, and it provides a useful dimension for improving advice. Conducting extensive research to determine how personalized teaching methods in FARM-HPT accommodate larger datasets and impact retention among students, outcomes for learning, and overall academic achievement.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
References
Xiang, D., Zhang, Z.J.: Cross-border e-commerce personalized recommendation based on fuzzy association specifications combined with complex preference model. Math. Probl. Eng. 2020, 1–9 (2020)
Asadi, S., Jafari, S., Shokrollahi, Z.: Developing a course recommender by combining clustering and fuzzy association rules. J. AI Data Min. 7(2), 249–262 (2019)
Zhou, L.N., Wang, C.X.: Research on recommendation of personalized exercises in English learning based on data mining. Sci. Program. 1, 5042286 (2021)
Xu, Y.: Computer-aided design of personalized recommendation in teaching system. Comput. Aided Des. Appl. 17(S1), 44–56 (2019)
Lalitha, T.B., Sreeja, P.S.: Personalized self-directed learning recommendation system. Procedia Comput. Sci. 171, 583–592 (2020)
Wang, R.X., Yang, W.H.: The design and application of smart classroom teaching mode in higher vocational education based on deep learning. In: 2022 International Conference on Education, Network and Information Technology (ICENIT), Liverpool, United Kingdom. pp. 66–70 (2022)
Li, Z.: Research on university ideological education platform based on data mining technology. In: 2021 4th International Conference on Information Systems and Computer Aided Education. pp. 254–258 (2021)
Jiang, J., Zeng, L.: Research on individualized teaching based on big data mining. In: 2019 14th International Conference on Computer Science & Education (ICCSE). pp. 56–59 (2019)
Pan, Y.Y.: Research on intelligent education method based on data mining of distance education system. In: 2021 International Conference on Intelligent Transportation, Big Data & Smart City (ICITBS). pp. 586–589 (2021)
Zheng, F.X.: Personalized education based on hybrid intelligent recommendation system. J. Math. 2022, 1–9 (2022)
Meddeb, O., Maraoui, M., Zrigui, M.: Personalized smart learning recommendation system for Arabic users in smart campus. Int. J. Web-Based Learn. Teach. Technol (IJWLTT) 16(6), 1–21 (2021)
Zhang, Z.G.: Personalized recommendation method of English micro-lectures teaching resources based on internet of things platform. Int. J. Inf. Commun. Technol. 20(2), 115–132 (2022)
Manogaran, G., Shakeel, P.M., Baskar, S., Hsu, C.H., Kadry, S.N., Sundarasekar, R., Muthu, B.A.: FDM: fuzzy-optimized data management technique for improving big data analytics. IEEE Trans. Fuzzy Syst. 29(1), 177–185 (2021)
Yuan, J., Chen, Z., Wu, M.: A novel distance measure and CRADIS method in picture fuzzy environment. Int. J. Comput. Intell. Syst. 16(1), 1–16 (2023)
Amin, F., Rahim, M., Ali, A., Ameer, E.: Generalized cubic Pythagorean fuzzy aggregation operators and their application to multi-attribute decision-making problems. Int. J. Comput. Intell. Syst. 15(1), 92 (2022)
Gao, P., Li, J.Y., Liu, S.: An introduction to key technology in artificial intelligence and big data driven e-learning and e-education. Mob. Netw. Appl. 26(5), 2123–2126 (2021)
Li, L.X., Huo, Y., Lin, C.W.: Cross-dimension mining model of public opinion data in online education based on fuzzy association rules. Mob. Netw. Appl. 26(5), 2127–2140 (2021)
Wang, T., Zhu, Y.F., Ye, P.J., Gong, W.C., Lu, H., Mo, H., Wang, F.Y.: A new perspective for computational social systems: fuzzy modeling and reasoning for social computing in CPSS. IEEE Trans. Comput. Soc. Syst. 11, 101–116 (2022)
Komleva, N.V., Vilyavin, D.A.: Digital platform for creating personalized adaptive online courses. Open Educ. 24(2), 65–72 (2020)
Fedushko, S., Ustyianovych, T., Syerov, Y.: Intelligent academic specialties selection in higher education for Ukrainian entrants: a recommendation system. J. Intell. 10(2), 32 (2022)
Dogan, O.: A recommendation system in e-commerce with profit-support fuzzy association rule mining (P-FARM). J. Theor. Appl. Electron. Commer. Res. 18(2), 831–847 (2023)
Papakostas, C., Troussas, C., Krouska, A., Sgouropoulou, C.: Personalization of the learning path within an augmented reality spatial ability training application based on fuzzy weights. Sensors 22(18), 7059 (2022)
Tran, T.T.T., Nguyen, T.N., Nguyen, T.T., Nguyen, G.L., Truong, C.N.: A fuzzy association rules mining algorithm with fuzzy partitioning optimization for intelligent decision systems. Int. J. Fuzzy Syst. 24(5), 2617–2630 (2022)
Modak, S.: Distinction of groups of gamma-ray bursts in the BATSE catalog through fuzzy clustering. Astron. Comput. 34, 100441 (2021)
Borlea, I.D., Precup, R.E., Borlea, A.B., Iercan, D.: A unified form of fuzzy C-means and K-means algorithms and its partitional implementation. Knowl. Based Syst. 214, 106731 (2021)
Cui, Y.: Intelligent recommendation system based on mathematical modeling in personalized data mining. Math. Probl. Eng. 1, 6672036 (2021)
Niknam, M., Thulasiraman, P.: LPR: a bio-inspired intelligent learning path recommendation system based on meaningful learning theory. Educ. Inf. Technol. 25(5), 3797–3819 (2020)
Rohidin, D., Samsudin, N.A., Deris, M.M.: Association rules of fuzzy soft set based classification for text classification problem. J. King Saud Univ. Comput. Inf. Sci. 34(3), 801–812 (2022)
Tutsoy, O., Koç, G.G.: Deep self-supervised machine learning algorithms with a novel feature elimination and selection approaches for blood test-based multi-dimensional health risks classification. BMC Bioinform. 25, 103 (2024). https://doi.org/10.1186/s12859-024-05729-2
Lin, M.W., Chen, Z.Y., Liao, H.C., Xu, Z.: ELECTRE II method to deal with probabilistic linguistic term sets and its application to edge computing. Nonlinear Dyn. 96, 2125–2143 (2019)
Zhang, J.P., Lin, M.W., Pan, Y.B., Xu, Z.S.: CRFTL: cache reallocation-based page-level flash translation layer for smartphones. IEEE Trans. Consum. Electron. 69(3), 671–679 (2023)
Lin, M.W., Li, X.M., Chen, L.F.: Linguistic q-rung orthopair fuzzy sets and their interactional partitioned Heronian mean aggregation operators. Int. J. Intell. Syst. 35(2), 217–249 (2023)
Lin, M.W., Huang, C., Chen, R.Q., Fujita, H., Wang, X.: Directional correlation coefficient measures for Pythagorean fuzzy sets: their applications to medical diagnosis and cluster analysis. Compl. Intell. Syst. 7, 1025–1043 (2021)
Lin, M.W., Li, X.M., Chen, R.Q., Fujita, H., Lin, J.: Picture fuzzy interactional partitioned Heronian mean aggregation operators: an application to MADM process. Artif. Intell. Rev. (2022). https://doi.org/10.1007/s10462-021-09953-7
https://www.kaggle.com/datasets/anlgrbz/student-demographics-online-education-dataoulad
https://www.kaggle.com/datasets/atulanandjha/lfwpeople?resource=download
https://www.kaggle.com/datasets/junyiacademy/learning-activity-public-dataset-by-junyi-academy
Funding
There was no funding.
Author information
Authors and Affiliations
Contributions
The author confirms being the sole contributor of this work and has approved it for publication.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that there are no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Li, D. Creating Personalized Higher Education Teaching System Using Fuzzy Association Rule Mining. Int J Comput Intell Syst 17, 239 (2024). https://doi.org/10.1007/s44196-024-00641-2
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s44196-024-00641-2