1 Introduction

The global economy has collapsed due to the coronavirus (COVID-19) pandemic [1, 22, 41]. New strain variants, a lack of social self-control, and optional vaccination all increase the likelihood that COVID-19 will persist and behave like a seasonal sickness. All nations are developing plans to gradually resume their economic and social activities since the socioeconomic situation has grown unsustainably [6, 79]. The COVID-19 pandemic prompted countries to impose strong limitations throughout 2020, portraying a scenario of decreased hospital visits that is unprecedented and the use of artificial intelligence (AI) and the internet of things (IoT)-based on the management of the pandemic. AI-based approaches to analyze, detect, classify and predict the trend of the deadly disease were developed [42].

Since its inception in 1956, AI has been studied to create “intelligent agents”—devices that can sense their surroundings and respond in ways that increase the possibility that they will succeed in attaining their objectives [35]. AI in healthcare started with the creation of expert systems, which were based on rules gleaned from expert interviews, and then translated and programmed [71]. This expert system involves the use of AI for the analysis, learning and deduction of inference from data. AI techniques have three methods, they are DL, ML and AI itself. The relationship between AI, ML, and DL is depicted in Fig. 1 [3].

Fig. 1
figure 1

AI/ML/DL Relationship

Full size image

The terms AI, ML, and DL are all somewhat confusing nowadays, DL is a subset of ML, which is a subset of AI, which is a catch-all phrase for any intelligent computer program. In other words, all ML is AI, but it is not all AI is ML and ditto DL. In the domains of computer science, data analysis, software engineering, and AI, ML and DL represent a significant evolution [10]. The study of AI is an active topic of study with many promising areas for further research including the detection and prediction using various techniques and applications [86]. Manual detection of pandemics and infectious diseases (IDs) is quite difficult because it takes a lot of time and is prone to human error. The potential of AI techniques is being further explored in the creation of automated and precise COVID-19 detection and prediction systems. Convolutional neural networks (CNN), in particular, have grown significantly in prominence among AI approaches for the detection, classification, diagnosis, prediction and other healthcare-related issues [27, 57, 84].

The continuing new Coronavirus (COVID-19) epidemic has caused millions of infections and many fatalities globally [56]. The World Health Organization (WHO) declared COVID-19 a global pandemic on 2020, the WHO is the only recognized body to declare IDs or any diseases to be a pandemic [45]. In the past, a lot of scholars have tried to forecast a COVID outbreak and its consequences [45]. Some people believe that time-series variables are the primary factors that can affect the onset of viral infections such as influenza and severe acute respiratory syndrome (SARS) [13, 67]. Worldwide, numerous IDs pose a threat to life. While certain viruses, like MERS, generate pandemics by moving from one country to another and across continents, others, like Ebola, circulate only in a small number of nations. Numerous studies have been done to quantify and forecast the occurrence of IDs as a result of the global epidemic of IDs [55].

IDs are infections that have suddenly developed in a community or that have existed but are now spreading quickly over the globe. IDs include the following: known infections spreading to new populations and areas; the revelation that an infection is the primary cause of a known disease; A new infection brought on by changes in the microorganisms, previously undiscovered infections emerging in locations whose habitat is changing; and an old infection resurfacing because it is no longer treatable or as a result of a failure in public health systems. When IDs transient a geographical region (country), it becomes a pandemic [61].

Rohr et al. (2019) [80] presented the links between Outbreaks of illnesses in humans and massive food distribution Infectious illnesses are spreading at an astonishing rate, wreaking havoc on global economics and public health. Thus, the social and environmental variables that give birth to microbial pathogens, as well as management techniques that may lower the likelihood of disease emergence or re-emerging, are of special importance [53]. At the same time, undernutrition which is a lack of one or more nutrients, continues to be a major cause of worldwide disease. Infectious illnesses are spreading at an astounding level over the world, while global food consumption is expected to skyrocket by 2100. The paper summarizes the methods through which anticipated agricultural growth and intensification will affect human infectious illnesses, as well as how human IDs may affect food production and distribution.

The study employs the systematic literature review (SLR) approach to study the trend of research in DL techniques for the detection and prediction of pandemics, SLR was chosen because it is based on well-defined methods and aims to find, assess, and aggregates all relevant evidence/ literature about the subject matter in a fair, repeatable and auditable manner. The following are the objectives of this survey:

  1. i.

    Study the trend of the pandemic since its inception globally;

  2. ii.

    Identify the technological innovations available through the use of AI techniques in the detection and prediction of pandemics;

  3. iii.

    Study the significance of DL techniques for detecting and predicting pandemics and;

  4. iv.

    Identify the limitations of the existing DL techniques in the detection and prediction of pandemics

  5. v.

    To ameliorate the open issues with the key strength of DL techniques for potential scholars to benefit from the open issues.

This research is organized as follows. Section 2 presents the technologies for the detection and prediction of pandemics, this includes data capturing technologies, pandemic data analysis and pandemic management technologies. Section 3 presents the survey methodology, while Section 4 contains the results and discussion of the survey, and Section 5 concludes the survey and presents future considerations to realize an efficient DL technique for the detection and prediction of pandemics.

2 Technologies for the detection and prediction of pandemic

Technologies play an important role in pandemics and IDs detection and prediction. The IoT can help by creating an early warning system to stop the spread of dangerous diseases. Integrated IoT networks, advancements in data analytics, AI, and universal networking on a worldwide scale are yet required to achieve this. These have been extremely beneficial for many aspects of humanity, especially in terms of preventing contagious diseases. A global network of IoT sensors in the healthcare sector offers both short-term and long-term benefits. Healthcare professionals and legislators have been able to follow any person who has been “compromised” as they pass through border controls. This would make it possible to focus on quarantine and, provide quick care, which would stop the coronavirus and other dangerous diseases from spreading. Long-term negotiations for the creation of a global early warning system should begin between large multinational organizations like the WHO and the United Nations. Such a tool could identify IDs before they spread globally. Worldwide emergencies, like the coronavirus, cause a number of fatalities, heightened stock price volatility and instability. A global detection system will clear up this uncertainty and give decision-makers the financial chance to react rapidly to pandemics and emergencies affecting all aspects of public healthcare [7].

2.1 Pandemic management technologies

Digital technology’s involvement in pandemic containment can take numerous forms and deliver great benefits. Thus, big data and AI research are linked to the pandemic that has plagued the world for decades (for instance the Spanish flu, SARS, MERS, avian flu, Ebola, and SARS-COV-2) as a fascinating and significant study stream. Along with the widespread use of digital health technology, another intriguing phenomenon relating to the collective activities done by the many players participating in the Healthcare Ecosystem may be seen in the management of the technologies for data collection, diagnosis, spread control and many more [50, 81]. Specialists, virologists, supranational agencies, ministries, hospitals, device suppliers, and all other entities working to minimize pandemic transmission are organizing their operations. While the necessary measures include striving to cooperate with quarantine, epidemic clusters, and country lockdowns owing to the lack of a viable particular vaccine. Blockchain technology, IoT, robotics and other related technology assisted have been helpful in the management of the ongoing pandemic [11].

The massive amount of complicated data from various sources about the number of patients, IDs contacts, mortalities, and so on has necessitated the use of big data technology to facilitate timely recognition of patient profiles, recurring and comparable trends of evolving disease actions, use and reaction to therapies, and so on. The use of big data in the administration of the pandemic, as well as the application of prediction algorithms, will be crucial in understanding the pace of viral transmission and the population at risk, as well as the natural history of the infection, high fatality rate, as well as the most effective treatment choices and preventative and control strategies [85].

Nguyen et al. (2020) [72] proposed a framework for COVID-19 detection based on data from smartphone sensors like webcams, microphones, heat, and sensor systems. The ML algorithm was employed in Italy, as well as many other nations throughout the world, to learn and gather information about illness symptoms based on collected data. In comparison to standard medical kits or skilled scanners, this strategy provides a low-cost and quick way for coronavirus screening since data inferred from sensing devices may be used effectively in a variety of specific applications.

AI technologies show significant effectiveness in assisting decision-makers in the virus management process. Allam and Jones (2020) [15] urged the use of AI and data-sharing regulatory mechanisms, to improve worldwide knowledge and control of urban health during the COVID-19 epidemic. For instance, when AI is combined with IoT devices deployed in many smart cities for early epidemic detection, further benefits can be realized. When medical data is gathered and distributed throughout and within smart cities. Rao and Vazquez (2020) [77] proposed a phone-based online questionnaire to collect people’s travel histories and common indicators. The acquired data may be evaluated using ML algorithms to study and predict the risk of infection, allowing for the early identification of high-risk cases for isolation. It limits the virus’s propagation to susceptible persons. In a recent review of DL detection research applications, challenges, and future directions by [68]. The study focused on the review of DL applications in healthcare featuring abdomen, cardiac, pathology, retina and diabetic retinopathy. The review did not consider the detection and prediction of pandemics.

2.2 Pandemic data capturing technologies

An innovative community-wide monitoring method called wastewater-based epidemiology (WBE) provides thorough real-time data on the state of public and environmental health and can support public health initiatives, such as those aimed at preventing IDs outbreaks. A potentially affordable method of keeping track of disease progression through WBE to stop local outbreaks is the use of Biosensors. In order to avert future pandemics, [52] reviewed the technical and economic viability of 18 recently developed Biosensors for the detection of IDs pathogens in wastewater. Municipalities lacking centralized laboratories, however, are probably still unable to analyze WBE samples.

Due to the paucity of reliable knowledge regarding the disease and its behaviors, detecting and predicting pandemic and IDs is a particularly difficult undertaking. Recent research has looked into several ways to prevent pandemics and infections effectively using IoT sensors. For the detection and prediction of pandemics and IDs, numerous researchers use the IoT to gather real-time sensory input data. One of [65] main investigations was a set of sensors that were dispersed across the workplace to look for or acquire information about pandemics and IDs. Real-time data could be captured by sensors and stored in the cloud, but the user would need to be made aware of the actual circumstances surrounding the data. The obtained data is subjected to filtering and analytics algorithms, which extract the data in the form of user information. Meraj, et al., (2021) [65] discussed COVID-19 as the most recent pandemic while also discussing the flu, zika, and H1N1. The research inquiry of the document also focused on the method of Remote Excess of Experts with IoT data.

2.3 Pandemic data analysis technologies

Various technologies were used to analyze pandemic data such as data mining, AI, ML, DL, IoT, and Blockchain among others. Kumar, Misra, & Chan (2022) [58] presented diagnostic technologies to analyze the coronavirus disease pandemic. In contrast to earlier coronavirus outbreaks of SARS and Middle East Respiratory Syndrome, chest X-ray has not shown adequate sensitivity to be useful in first-line screening techniques. Despite the fact that current national and international guidelines propose using reverse transcription polymerase chain reaction (RT-PCR) as the initial screening technology for identified COVID-19 infections, academic and regional procedures must reflect local resource allocation when releasing general guidelines. So far, successful containment and social mitigation techniques have relied on united governmental reactions, even though the underlying philosophies of these activities may not be universally applicable in many Western nations. As the need for radiology personnel grows, preliminary findings point to a prospective role for machine-learning techniques as risk prediction criteria.

Pan et al., (2020) [75] opined during the recovery from the coronavirus pulmonary outbreak, studying the temporal course of lung alterations on chest CT was studied. Patients with RT-PCR proven COVID-19 infection were included in the retrospective study. Patients who experienced significant respiratory distress and/or required oxygen at any point throughout the disease’s progression were excluded. Repeat chest CTs were acquired at 4-day intervals. The overall CT score was calculated as the sum of the lung activity indicated. Twenty-one individuals with COVID-19 pneumonia were assessed. These patients had a total of 82 lung CT scans and were all released after an average of 17 days in the hospital. The degree of lung irregularities on chest CT in the individuals recovering from COVID-19 pneumonia was greatest roughly 10 days following the beginning of symptoms.

Chakraborty et al. (2020) [32] presented a DL classifier sentiment analysis of COVID-19 pandemic tweets. People are fearful of panic, terror, and worry as the number of instances increases exponentially over the world. The physical and mental health of the global populace has been shown to be inversely correlated with this pandemic illness. The study stressed the fact that tweets comprising all handles associated to COVID-19 and WHO have been ineffective inappropriately advising people around this pandemic epidemic. The study indicated that, despite the fact that individuals tweeted primarily nice things about COVID-19, netizens were preoccupied with retweeting the bad tweets, and that no beneficial terms could be located in WordCloud or calculations based on word frequency in twitter posts. The claims were confirmed using a suggested model that included deep learning models with allowable accuracy of up to 81%. Aside from this, the researchers have suggested the use of a fuzzy rule based on a Gaussian membership function to accurately identify attitudes in tweets. The accuracy of the aforementioned model is up to an acceptable rate of 79%.

Oh, et al. (2022) [73] applied in low COVID-19 occurrence circumstances, neighborhood-scale wastewater-based epidemiology was used. WBE is a new technique for neighborhood COVID-19 monitoring, it has been studied largely at major sewer sheds such as wastewater treatment facilities that serve a massive community. From January to November 2021, WBE was applied to seven neighborhood-scale sewer sheds. Data from society testing revealed a mean occurrence of 0.004% in these sewersheds (97.0% of monitoring periods indicated two or less daily illnesses). The research findings indicated that neighborhood-scale WBE can detect local epidemics that city-scale WBE would miss. As a result, the author’s findings indicated that neighborhood-scale WBE was an effective community-wide disease surveillance technique while pandemic incidence remains low.

Zheng et al. (2022) [91] presented influencing variables and coronavirus clustered features. The study used descriptive statistics and regression analysis on COVID-19 data from several nations to analyze and assess various regression models. The extreme random forest regression model performed best, and parameters like population size, ozone, average age, life span, and human development index were found to have a moderate influence on the dissemination and dispersion of COVID-19 in the ERFR-based method. The spectral clustering technique was used to examine the pandemic clustering properties. The spectrum clustering visualization findings revealed that the geographical distribution of worldwide pandemic spreading creation was highly clustered, and its grouping features and impacting variables also demonstrated some uniformity in dispersion.

Oyelade, & Ezugwu (2020) [74] presented a case-based reasoning approach for new coronavirus identification and diagnosis. The study applies an upgraded CBR model for cutting-edge reasoning tasks in the categorization of suspected COVID-19 instances. For case similarity calculation, the CBR model makes use of a unique selecting features and the semantic-based computational formalism introduced in this paper. The database was first populated with 71 cases received from the Italian society of medical and interventional radiology (SIRM) repository. The results showed that the suggested technique in this study correctly sorted reported cases into respective categories with a 94.54% accuracy. The study discovered that the suggested approach can assist clinicians in readily diagnosing probable cases of COVID-19 based on patient medical records without the samples being sent to laboratory testing.

2.4 Deep learning for detection and prediction using COVID-19 dataset

The manual detection of the infection using radiographic pictures is quite difficult since it takes a lot of time and is very prone to human error [28]. The development of automated, precise, and effective methods for pandemics and Identities, including COVID-19 detection, has showed promise and is now being pursued. For the categorization of COVID-19, DL-based strategies have been increasingly prominent among AI technologies [6]. The proposed system included three stages: preprocessing with lung segmentation, removing the environment that did not provide relevant information for the task and could lead to biased results; classification model trained under transfer learning scheme; and finally results analysis and interpretation via heat maps visualization. The most accurate models attained a 97% COVID-19 detection accuracy [23].

A recent study suggested an automatic CNN-based DL-based classification technique that showed a quick COVID-19 detection rate. 3616 COVID-19 chest X-ray pictures and 10,192 images of healthy chests that were later enhanced make up the training dataset. The COVID-19 symptoms were initially identified utilizing the dataset by applying eleven pre-existing CNN models. The potential of MobileNetV2 made it a candidate for further improvement. Of all the deployed CNN models, the final model produced the greatest accuracy of 98% in classifying COVID-19 and healthy chest X-rays. The findings imply that the suggested strategy outperforms existing ones in accurately detecting illness symptoms from chest X-ray pictures [12]. A DL-based method to categorize COVID-19 based on x-ray images was recently introduced in a study. The results were encouraging and might be used to distinguish infected individuals from healthy individuals. The Kaggle dataset of COVID-19 X-ray pictures was used for the trials, which were carried out using a ResNet50 DL-based approach with 5 and 10 folds cross-validation. The experiment’s findings demonstrated that 5 folds had an accuracy rate of 97.28%, which was higher than 10 folds [4].

A recent study examined and contrasted various DL-based enhanced approaches used to detect COVID-19 in X-ray and CT-scan medical pictures. VGG16, DenseNet121, ResNet50, and ResNet152 were four pre-trained CNN models that were employed for the binary classification job for the COVID-19 CT-scan. The appropriate architecture, pre-processing, and training settings for the models will be determined largely automatically via the Fast.AI ResNet framework. The F1-score and accuracy for the diagnosis of COVID-19 using CT-scan images were both over 96%. In order to overcome the lack of data and shorten the training period, we also used TL-based techniques. X-ray image tasks were classified in binary and many classes using the modified VGG16 deep transfer learning architecture [90].

3 Methodology

This section describes the process used to locate articles and how those articles were used to determine the specified objectives in relation to the selection criteria and the solutions to the research questions. The four goals of the current survey were established through four separate research questions (RQ) [17]. Figure 2 shows the selection chart and criteria for research surveys. This section proffers the review guide that anchored SLR principles in this survey.

  • RQ1: What has been the pandemic experience globally?

  • RQ2: How far has AI/DL/ML techniques contributed to the detection and prediction of pandemics?

  • RQ3: How has DL techniques contributed to the detecting and predicting pandemics?

  • RQ4: What are the limitations of the existing DL techniques in the detection and prediction of pandemics?

Fig. 2
figure 2

The chart for analyzing research findings and the selection criteria

Full size image

3.1 Search techniques

The references for this review were discovered through Scopus searches. This comprises journal articles, chapters from conferences that were referred, and other types of content from top research databases that Scopus indexed between 2013 and 2022. Because they were regarded as trustworthy databases for study in AI, computer science, and biomedical sciences in general, these databases were chosen. These include Sensors, IEEE Access, Computers in Life and Medicine, Electronics Switzerland, Computer Materials a Continua, and Applied Sciences. We incorporated items from January 2013 through September 2022 that were found using these searches.

3.2 Study selection/ studying choice

Both prospective and retrospective studies that reported on original English-language research on the detection and prognosis of pandemics were taken into consideration. Reports on the development and application of DL approaches for pandemic detection and prediction are available. To concentrate on the research work with in-depth knowledge of the study area and studies with the detection and prediction of pandemics using information not currently available in routine clinical care or with results unrelated to the typical care provided by physicians, conference papers, journal articles, and chapter contributions were included, with the exception of one. Others were excluded based on our selection criteria.

3.3 Data analysis in the selected articles

Articles were first chosen based on sections 4.1 and 4.2. Three concepts (detection, prediction, and pandemics) appeared in the titles, abstracts, and keywords of the publications we obtained. Each abstract, type of pandemic/dataset, methodology, and performance evaluation used, as well as the outcomes obtained, were researched and evaluated. Publications using DL techniques were carefully examined and analyzed. 44 papers from the final portfolio were subjected to a bibliometric and content analysis, which grouped the research’s findings and observations.

3.3.1 Inclusion criteria

Published studies in English no sooner than January 2013 and no earlier than October 2022. Research Gate, Semantic Scholar, PubMed, Google Scholar, and Science Direct published databases.

3.3.2 Exclusion criteria

Types of published papers (review, research, letter and journals). The search did not include the reviewed or non-DL papers.

Duplication: Duplicated papers were excluded from the search results.

Time of publication: The time frame for the published research was ten years, from 2013 to 2022.

Language of publication: English was employed as the study’s official language and as a qualification for inclusion.

Process for the Research: From 2013 to 2022, the research was conducted by hand-searching certain journal articles. The following research methodology was applied to this review:

  1. i.

    Resources: The process for the research includes journal articles published between January 2013 and October 2022. In addition, papers that used DL for the detection/prediction of pandemics in the last ten years (2013-30th of September 2022) among these were further investigated. Thus, references of all selected journal articles were checked to see if pertinent papers had been selected.

  2. ii.

    Research Keywords: The following keywords were utilized in the research questions for this review: Detection, prediction and pandemic.

  3. iii.

    Research Expressions: For this review, the guidelines for utilizing the research words have been followed: Keywords terms are taken from research questions related to the pandemic using DL techniques. The AND logical operator was used to group keywords into categories, and the OR logical operator was used to combine phrases and synonyms.

4 Results and analysis

4.1 General outcome of the literature search

A total of 790 research papers were extracted using the stated search terms. The statistics of these papers are presented in Table 1, which covers the journal article, conference papers, chapters in books, letters editorials and so on. Each of the paper titles was examined independently using the inclusion and exclusion criteria, leaving 44 papers that either work on pandemic detection or prediction or both. 746 publications were deleted because they had no relevance to the research subject (some were excluded because they were focused on other aspects of detection and prediction non-specific to the pandemic, some are pandemic studies but they are bot detection and/or prediction). Before these 8320 were detected, they were further reviewed based on their abstracts, introductions, and conclusions and found none relevant to the subject area. 4 duplicate papers were removed using Endnote X8. Several articles were eliminated because they were difficult to understand in their entirety or because their abstracts showed that they had no bearing on the question. A total of 44 papers were selected for journal article assessment and state-of-the-art review; each was examined in its completeness, separately, and again using the inclusion and exclusion criteria. Figure 3 shows the graphical representation of the number of publications received and published before and after the COVID-19 pandemic. The graph shows that the highest publications on the pandemic were published in the year 2021. Figure 4 presents the most frequent word led by humans and followed by COVID-19, DL also appears making it a prominent AI technique that is contributing to the detection and prediction of pandemics.

Table 1 Classification of Publication by document type
Full size table
Fig. 3
figure 3

Trend of the detection and prediction of pandemic publications

Full size image
Fig. 4
figure 4

Most frequent words

Full size image

4.2 Outcome of our findings based on research questions

Some of the selected articles uniquely answered the questions by addressing one unique facet of Pandemic detection and prediction based on search area, application area, and geographical area among others. While other articles mostly re-emphasized what was written in other articles.

4.2.1 The global trend of the pandemic

RQ1: What has been the pandemic experience globally

Presenting the statistics from the evaluated literature, the global trend of pandemic based on the review literature and using selected search terms in Scopus in October 2022 is the focus of Fig. 5 showing the trends of the pandemic, Saudi Arabia, United States, China, Pakistan and Iran are top five (5) countries with cases of pandemic globally. Conversely, Kasakhstan, Mexico, Palestine, Peru, and Ukraine are at least five (5) countries with cases of pandemic globally. It is worthy of note that the studies on DL techniques and prediction increased geometrically between 2019 and 2022, this may not be unconnected to outbreak of COVID-19 pandemic. Meanwhile, many countries in Africa that has no sophisticated healthcare infrastructure needs to prepare for eventualities, many developed countries were able curtailed the pandemic because of their sophisticated healthcare facilities. These answered (RQ1). To further explore the pandemic experience globally, the most cited countries in the reviewed publications, was investigated and the outcome is presented in Fig. 6 as the most cited countries. The United State of America is leading followed by China.

Fig. 5
figure 5

Trends in the number of publications using author country as a search key between January 2013 to September 2022

Full size image
Fig. 6
figure 6

Ten most cited countries

Full size image

In answering RQ1, the pandemic, especially the COVID-19 pandemic continues to impact all aspect of lifes which is affecting 288 countries and territories globally. COVID-19 as one the pandemic that has affected the world infected over 626,360, 380 people globally while 6,560,192 deaths have been recorded as of 8th October 2022 [89]. The pandemic has disproportionately hurt the weak and the poor, and it poses a threat to further impoverish millions of people. The pandemic caused a lockdown and shutdown of global business which resulted in the current high inflation rate in the global market. In terms of publiccation, major attention was shifted to the COVID-19 pandemic which cause the rapid increase in COVID-19 articles in the last three (3) years, although the present study covers ten years of the survey.

4.2.2 Technological innovations available for the detection and prediction of pandemics

RQ2: How far has AI/DL/ML techniques contributed to the detection and prediction of pandemics?

Presenting the statistics of the literature reviewed, the trends in the number of publications using selected search terms in Scopus in October 2022 were the focus, Fig. 7 shows the comparison of the trend of using the three AI-based technologies publication in pandemic using selected search terms in Scopus for the period under review. These answered research question 2 (RQ2). In the analysis of technological algorithm in detection and prediction of pandemic over the period of ten (10) years. It comprises of AI/ML/DL techniques, DL techniques were the least AI-based technological used for pandemic analysis and studies, it was reported in 3573 publications, most of these publications were not focusing on the detection and prediction of pandemics. Figure 7 gives the details of the contributions of AI-based studies extracted from the database, it presents the comparison of three AI-based technological tools used in the analysis of pandemic data the last 10 years. Figure 8 presents the statistical chat of three AI-based technologies for COVID-19 pandemic detection and prediction over the period of 10 years in the current study.

Fig. 7
figure 7

Comparison of the trend of using the three AI-based techniques used in pandemic

Full size image
Fig. 8
figure 8

Statistical Chat of Three AI-based techniques for COVID-19 Pandemic Detection and Prediction Over the Period of 10 Years

Full size image

4.2.3 Significance of DL techniques detecting and predicting pandemics

Although, DL was the least AI-based used among the three. This calls for further investigations. Hence, the need for RQ3, that inquires for the significance of DL techniques for detecting and predicting pandemics. DL is significant in the detection and prediction of pandemics, it was mentioned in 1840 publications and 11 fields or subject areas.

RQ3: How has DL techniques contributed to the detecting and predicting pandemics?

The analysis of DL-based state-of-the-art studies reviewed shows the novelty DL-based techniques in the pandemic detection and prediction. DL techniques were used for various purposes in the studies including the detection and prediction of pandemics. Table 2 presents the outcome of the state-of-the-art reviewed literature on DL for the detection and predictions.

Table 2 State-of-the-art Review of DL Studies Focusing on Detection and Prediction
Full size table

DL-based techniques in contributing in no small measure to the pandemic control and healthcare delivery as a whole, it facilitates the detection, prediction, diagnosis and medical treatments of the infectious persons. Also, improvement in the applications for screening, recognition, segmentation, and classification across numerous areas of healthcare have been made possible by the use of DL techniques [68]. The results from the state-of-the-arts review indicates that, the use of optimization and feature selection approach was substantially efficient in terms of various performance in comparison to existing models DL techniques, but yet to be adopted in the detection and prediction of pandemic.

4.2.4 Limitations and challenges of DL techniques in the detection and prediction of pandemics

RQ4 What are the limitations of the existing DL techniques in the detection and prediction of pandemics?

The success of DL technology did not come without some challenges such as low detection and classification accuracy, which is a problem with pandemic detection and prediction [49]. Incomplete, improper label, unbalanced and biased dataset [78]. In addition, global challenges for the development of ML/DL/AI for clinical decisions are to be considered. Table 3 describes the main challenges for the detection and prediction of pandemics using the existing DL technology that have been noted in the literature. It is significant to note that each study in the literature has demonstrated the potential for automated pandemic detection and prediction but has also encountered difficulties or lacked thorough investigation and evaluation of the suggested solutions from many angles. It is worthy of note that COVID-19 pandemic dataset was the most prominent pandemic dataset that DL techniques were used to implement within the year under review, hence, the current survey is limited to COVID-19 pandemic studies.

Table 3 Limitation of DL techniques in the detection and prediction of pandemics
Full size table

Different factors need to be considered to successfully implement the detection and prediction of a pandemic using DL techniques. In order to inform the research community and help it create more effective, reliable, and accurate DL models for the various challenges in the detection and prediction of pandemics, a critical analysis of the pertinent studies and techniques is conducted. The associated limitations are highlighted, and the research gaps and future challenges are identified. In order to inform the research community and help it create more effective, reliable, and accurate DL technique for the various limitations in the detection and prediction of pandemics, a critical analysis of the pertinent studies and techniques is conducted. The associated limitations are highlighted, and the research gaps and future challenges are identified.

4.3 Outstanding issues in using deep learning solutions in pandemic diseases

DL-based techniques and architecture nevertheless have certain practical restrictions or downsides or gaps in their usage to predict and detect future pandemics. This paper highlights the limitations of DL-based techniques in the detection and prediction with respect to pandemics as scope for future research directions and the outstanding issues. These gaps include poor data quality management, difficulties in interpreting detection and prediction outcomes and low accuracy to mention but a few. One of the key findings of this survey is that most of the state-of-the-art studies did not apply optimization techniques, and/or feature selection in their studies, and few studies that applied either optimization or feature selection techniques did not predict of future occurrence of pandemics in the study. These issues may be addressed by developing hyper-parameter-based using bioinspired optimizer of DL techniques that modify the networks structures and the training process, ultimately develops DL-based architecture to give reliable first-hand detection and prediction of pandemics. The DL architecture’s parameters can be optimized using pre-trained model and this can also be compared with non-trained DL architectures. Additionally, by highlighting the crucial features, dataset improvement approaches can help increase detection accuracy. Overall, researchers, engineers, computer scientists and healthcare experts who working in the field of AI/DL/ML will find the results of this paper useful for development of an effective techniques for detection and prediction of pandemics to further prepare for future pandemic and IDs thereby helping healthcare delivery systems.

5 Conclusion and future consideration

The previous and recent COVID-19 pandemic has inspired scientists to use various learning algorithm techniques to detect and predict the future occurrence of the pandemic. The current situation calls for a more accurate, more efficient, less complicated and inexpensive system that is capable of both the detection and prediction of pandemics and IDs. Although there are numerous diagnostic techniques for identifying pandemics such as SARS-CoV-2 infection, DL techniques happen to be one of the most widely used AI technologies to battle this pandemic. It has contributed in no small measure in curbing the pandemics as well as prevent its spread. This paper intends to find out various DL-based techniques and their contributions towards the detection and prediction of pandemics, Furthermore, this paper presents the state-of-the-art review of research activities about how AI/DL/ML techniques have contributed to the detection and prediction of pandemic in terms of pandemics monitoring systems, classification human activity recognition, data fusion, collecting vital signs of patients among others. In this survey, we reviewed more forty-four (44) published research papers written in a decade from 2013 to 2022 were reviewed. In order to understand the significance of DL contributions and limitations for the pandemic control. Hence, the current studies has contributed to the body knowledge by providing a comprehensive overview of the current state-of-the-art research in the field of AI particularly DL-techniques techniques for the detection and prediction of COVID-19 pandemic, identify gaps in the existing literature, and provide guidance for future research directions. This survey addresses and identifies a vacuum in the field by summarizing and assessing existing research that applied DL techniques for the detection and predictions of pandemics in their publication and identifies gaps in the existing literature for the purpose of future research direction in this aspect. An analysis of the different studies based on DL techniques for the detection and prediction of pandemic has been performed. The primary goal of this study is to provide researchers with some crucial research briefings that may help them create more effective and robust DL-based approach, that will be efficient and effective for the detection and prediction of pandemic. In order to explore the advantages of DL techniques better in the area of detection and prediction of pandemics and IDs control and prevention, this study considered varying challenges and alleviates them in different aspects such as feature selection, recognition, optimization and computational complexity.