Machine Learning and Internet of Medical Things for Handling COVID-19: Meta-Analysis

Early diagnosis, prioritization, screening, clustering and tracking of COVID-19 patients, and production of drugs and vaccines are some of the applications that have made it necessary to use a new style of technology to involve, to manage and deal with this epidemic. Strategies backed by artificial intelligence (AI) and the Internet of Things (IoT) have been undeniable to understand how the virus works and try to prevent it from spreading. Accordingly, the main aim of this survey article is to highlight the methods of ML, IoT and the integration of IoT and ML-based techniques in the applications related to COVID-19 from the diagnosis of the disease to the prediction of its outbreak. According to the main findings, IoT provided a prompt and efficient approach of following the disease spread. Most of the studies developed by ML-based techniques for handling COVID-19 based dataset provided performance criteria. The most popular performance criteria, is related to accuracy factor. It can be employed for comparing the ML-based methods with different datasets. According to the results, CNN with SVM classifier, Genetic CNN and pre-trained CNN followed by ResNet, provided Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 7 February 2022 doi:10.20944/preprints202202.0083.v1 © 2022 by the author(s). Distributed under a Creative Commons CC BY license. Machine Learning for COVID-19 2 highest accuracy values. On the other hand, the lowest accuracy was related to single CNN followed by XGboost and KNN methods.


Introduction
The COVID-19 outbreak has created many challenges in human life around the world [1,2]. The most devastating impact which is increasing casualties and deaths (around the world) have made it clear the need for social and business restrictions [3]. With the expansion of the COVID-19 pandemic, the world community has faced many other problems in various aspects of life such as, economic and social life, psychological wellness, political interactions, cultural activities, educational limitations, religious restrictions, and even sport events [4,5]. Such examples highlight the need for effective and intelligent systems to deal with such crises in pandemic situation [5]. Early diagnosis, prioritization, screening, clustering and tracking of patients, and production of drugs and vaccines are some of the applications that have made it necessary to use a new style of technology to involve, to manage and to deal with this epidemic [6]. Machine Learning (ML) and Artificial intelligence (AI) algorithms displayed promising ability in prediction and classification [7][8][9][10][11][12][13][14][15][16][17][18] including disease prediction [19][20][21][22][23][24][25][26][27][28][29][30], virus genome analysis [20,31,32], and medical imaging and Internet of Things [33][34][35][36]. Strategies backed by artificial intelligence (AI) and the Internet of Things (IoT) have been undeniable to understand how the virus works and try to prevent it from spreading [5,37]. These techniques have evolved with the development of computing resources with cloud computing and recent advances in machine learning (ML). All of these advances enable researchers to successfully process large amounts of data and extract information from it. ML-based methods used in processing and modeling data on COVID-19 disease can increase efficiency and speed up results by improving computations.
Gou et al. developed a survey for evaluating the ML-based techniques for diagnosis of COVID-19 using medical data collection, image preprocessing, feature extraction, and image classification. The study evaluates Transfer, ensemble, unsupervised and semi-supervised learnings, and convolutional neural networks, graph neural networks, and explainable deep neural networks. Evaluations focused on the advantages and limitations of the diagnosis techniques [38]. Abumalloh et al. presented a stateof-the-art of ML-based techniques for handling medical image processing in the context of the COVID-19 crisis [39]. Khan et al. developed a survey of the applications of AI for preventing the COVID-19 pandemic [40].
As can be deduced, many survey studies have been developed in this regard. But, the existence of a study that can systematically review and discuss two interrelated areas of the ML and the IoT in the form of an article has been lost from the research literature. Accordingly, the main purpose of this review article is to examine the methods of ML, IoT and the integration of IoT and ML-based techniques in the applications related to COVID-19 from the diagnosis of the disease to the prediction of its outbreak.
The study has three main sections. The first section is to present the importance of IoT and IoT-ML based techniques in COVID-19 applications. The second section is to present the importance of MLbased techniques in COVID-19 applications. The last part is to present the main findings, challenges and future perspectives.
Systematic review may successfully provide technical and effective literature for a specific topic [41]. A systematic review requires proper collection of papers around the subject. Preparing dataset is one of the main steps of determining the quality of review work [41]. Finding the most related studies requires proper keywords and proper searching libraries. In the present study, the most relevant studies have been collected from WOS, and Scopus libraries. Table 1 presents the main keywords and searching queries.

IoT for COVID-19
IoT is an interconnected set of computing tools from simple to complex that can be used in conjunction with mechanical or digital machines in the presence of humans, animals, or objects. IoT technology can easily transfer data from the source to the destination through the network without the presence of the operator. This technology can actually be considered as a special tool in human-human interaction or human-computer interaction [42,43]. An IoT platform includes the minimum equipment required, such as smart devices equipped with the web [44]. These systems consist of processors, sensors, and communication hardware to collect, send, control, manage, and generally convert data into accessible data [44,45]. These systems connect to an IoT port used to send data to the cloud so that data can be analyzed and shared [46]. These devices can operate by connecting to other related systems based on the information they receive [47]. These tools perform many of their tasks without human intervention.
Today, the use of IoT technology in the field of health and treatment is growing, rapidly [48]. The main applications of IoT in the process of treatment and intelligent health can include identification, digitization of medical information, patient transfer to the hospital, use of vital signs sensors, use of smart phones in communication, and in general, digitization of medical processes [49,50]. Use of IoT has become more popular and important due to COVID-19 pandemic [51]. Due to the fact that this virus is highly contagious and has a high risk to human health and has caused many problems for the medical staff, using non-contact methods to diagnose as soon as possible, control patients, monitor the condition of patients with acute illness, as well as maintaining social distance, can be an important factor in breaking off part of the virus infection cycle [52,53]. In the field of non-contact methods, the IoT is a leader and can solve many problems in this field [54]. Figure 1 presents the main applications of the IoT in COVID-19 era.
IoT in Covid-19  Table 2 presents the highlighted studies for the application of IoT-based techniques to tackle COVID-19. This table represent the studies based on the main four columns. Objective that briefly describes the main objective of each study. Methodology/proposed algorithm that presents the main algorithm and procedure employed by each study. Keyword indicates the main points and concentration of the study and finally, application section presents the field of the application of each method.     [70]. Figure 3 presents the main contribution of these papers. According to the reviewed studies, COVID-19 dataset can be imported from three main sources including Radiography, statistics of health centers, and Sensors for prediction, monitoring, identification, detection, diagnosis and classification purposes. The output of the techniques needs to be evaluated for confirming the approach performance and accuracy values. The frequently used parameters for performance analyzing include Accuracy, Precision, Recall, RMSE, Correlation coefficient and mean absolute percentage error. This can be considered as a brief explanation as the main contribution of the present study. This study successfully presents the advantages and disadvantages of each technique for a specific task in handling COVID-19 dataset and propose the future perspectives. Also, this study can detect the main challenges and limitations. There is a need to categorize the main applications of IoT and the relevance technique in accordance with the COVID-19. Table 3 presents the main contributions of the study for the application of IoT and integrated IoT-ML based techniques.

Radiography
As shown in Table 3, IoT based technology requires using ML-based techniques for completing the task. Figure 4 presents the share of each methodology in the applications by percentage. As shown in Figure 4, IoT has been used more than other applications to monitor and detect COVID-19 cases. While, it has been less popular in the identification.

ML techniques for pandemic prediction of COVID-19
Utilizing ML platform led to reducing the adverse effects of the disease and accelerating the healing process [51]. Combination of AI and ML have led to advances in treatment, medication, screening, prognosis, contact tracking and the drug/vaccine development process and reduced human intervention in medical performance [71]. ML is also used as a tool for management of virtual queues to prevent crowds in physical waiting rooms or long queues. It is used to predict waiting times and implement calls in a privacy manner in conjunction with the cell-phone platform [72].
ML method is widely used in data analysis by producing an analytical model intelligently. This method is a subset of artificial intelligence that analyzes data and produces a model for estimating, categorizing, optimizing, predicting, identifying problems, as well as decision making [73,74].
New computing technologies have made the problems assessed by ML-based techniques today a little different from the way they are analyzed based on past technologies [75]. These techniques began to evolve from pattern recognition to a comprehensive theory of the ability of computers to perform specific tasks without the need for special planning [76,77].
In the field of medicine and treatment, ML is known as one of the most practical tools for analyzing medical data, identifying, predicting, and even treating in different situations. With the advancement of medical science in today's world and the production of large volumes of medical data, there is an urgent need to analyze this data [78]. Figure 5 presents the main applications of ML-based techniques for medical science to tackle COVID-19 pandemic. Identifying the prevalence, effective parameters in the eradication of the virus, identifying patients in the early stages, patients' pattern behaviors, and predicting outbreak and mortality rates can be considered practical and effective areas of ML-based techniques [1,79].
Machine Learning Figure 5. the main applications of ML-based techniques for medical science Table 4 presents the highlighted studies for the application of ML-based techniques for handling COVID-19. Similar to Table 3, Table 4 discuss them in four columns. Objective column that briefly describes the main objective of each study. Methodology/proposed algorithm presents the main algorithm and procedure employed by each study. Keyword indicates the main points and concentration of the study and finally, application section presents the field of the application of each method. According to Table 4, ML-based techniques are employed for detection, identification, monitoring, diagnosis, prediction and classification purposes in the presence of the COVID-19 dataset. Figure 6 presents the summary of each application, separately. Singh and Kaur employed ML based platform using hybrid random forest, Gaussian Naïve Bayes and Generative adversarial network as a healthcare application to detect COVID-19 cases [81]. Vinod   According to Figure 6, detection, diagnosis, and prediction can be considered as the main categories of the application of ML-based methods in COVID-19. In general, one of the main sections of analyzing IoT-based and ML-Based techniques applied for a specific field is their evaluation in terms of accuracy, error, or in other word performance of the model. Table 5 presents the evaluation criteria employed for each model.
According to Table 5, accuracy, followed by recall and precision parameters have owned the highest portion of the evaluation criteria employment for analyzing COVID-19 based dataset using IoT and ML-based techniques. In the following, Table 6 is generated from Table 4 for indicating the share of each ML-based techniques for each application and their main contributions. According to Table 6, ResNet as an architecture of deep learning methods followed by CNN, XGBoost, SVM, DT and LR have been used more often to tackle work with COVID-19 related data.  Figure 7 presents the share of different ML methods for different tasks to tackle COVID-19 pandemic.
As it is clearly indicated in this figure, ResNet followed by CNN are the most common application of ML in this field. This can be due to the model's nature for handling different applications like monitoring, detection, identification, classification, and diagnosis. While, other methods are capable to do limited number of applications.

Evaluation criteria
Models developed using ML and IoT-ML requires an evaluation step for recognizing their performance and accuracy values. According to the studies reviewed, the most effective and frequently used evaluation criteria are including Accuracy, Recall, Precision, Root mean square error (RMSE), Correlation coefficient and Mean absolute percentage error (MAPE). These criterias compare the models' output and actual values and provides a comparison score [1,79]. In the present study, we employed the criteria values reported by each study for evaluating and comparing the models. Table 7 presents the main criteria for evaluation.    where denotes the total number of samples, the actual samples, and ̂ the predicted samples. (4)

= ( ,̂)
where refers to actual samples, ̂ to predicted samples, ( ,̂) to the covariance between and ̂, and σ to the standard deviation (calculated for both and ̂) where denotes the total number of samples, the actual samples, and ̂ the predicted samples.

Main findings and evaluations
This section presents the main findings of IoT based techniques (Table 8) and ML-based techniques (Table 9). Each table includes two main columns called findings and pros. and cons. 3 The model provides an accuracy of 98% for detection Combining DL and the IoT makes it easier for radiologists to control the spread of the virus [57] 4 According to results, all the techniques, except the Decision Stump, OneR, and ZeroR provided accuracies values more than 90% The proposed platform reduced the communicable diseases using early detection of cases and provided tracking the recovered cases, and a better understanding of the infections [58] 5 IoT reduces clinical cost and optimizes treatment outcome of the patients The platform improves patient satisfaction and decreases readmission rate in the hospital [59] 6 The system can assist tracking the daily activities and decrease the risk of exposure to the COVID-19 The app announces the user to keep a physical distance of 2m. Also, a Fuzzy-based technique evaluates the environmental risk and user health to estimate the risk of real time spreading. This platform can successfully reduce the coronavirus spread [60] 7 The platform detects and tracks the infected person The platform tracks COVID-19 and improves infected person and keeps the dataset for further analysis [61] 8 The provided package enhances the testing process for increasing the efficiency of the system This approach will increase the maximum collaboration from the employees This platform is a cost-effective, safety-critical mobile robotic technology and successfully copes with diagnosis task Also the multiple diagnostic devices increases the detection accuracies The system effectively provides a complete diagnosis and figuring out COVID-19 patients also contains multiple diagnostic devices, without any need for human interferences [63] 10 The robot technology protect virus affected persons.
The system is also recognizing the patient's Gesture and tracking the instructions The robot collects data from patient performs tasks without image processing system [64] 11 IoT-based technology prevent the global pandemic improves the control and tracking of a fast-spreading virus such as coronavirus [65] 12 The proposed methodology is sustainable for disease tracking by an early identification of cases This technique can successfully handles both governments and other decision-making authorities [66] 13 This system improves the decision-making procedure The system is connected through cloud computing and effectively supports the real-time data [67] 14 Edge computing improved the findings on the decentralized load of face recognition The platform enhances the robustness of detection and diagnosis [68] 15 The proposed system could successfully cope with the task IoT equipped ML can successfully save, and visualize monitoring the volunteers [69] 16 This study suggests that integrated and hybrid techniques will follow up the near future, using simulation, and forecasting purposes A higher degree of safety and privacy for humanity [34] 17 The platform employed for the study have an effective role in the success of pandemic handling The platform increases accessibility to the proper dataset [70] According to Table 8, most of these studies lack numerical analysis for the method performance. One of the main reasons can be the nature of IoT technique which goes through a practical process and shows its performance in practical applications and does not need to provide numerical statistics. In all these applications, IoT could successfully cope with the task. Such that, IoT provided a fast and efficient approach of tracking the disease spread [55]. On the other hand, it can be employed as a realtime framework to minimize the impact of communicable diseases through early detection of cases [56]. In the study by Singh et al , IoT technology successfully increased patient satisfaction and reduces readmission rate in the hospital [59]. However, for detection purposes, there is a need for integrating IoT platform with ML-based techniques. In the study by Rahman et al, DL applications with IoT platform provided promising findings to detect AE attacks. However, there is a need for further research, attention, and implementation of appropriate defense mechanisms, safeguards, and controls [56]. Kolhar et al employed Multi-task Cascaded Convolutional Network architecture (MCCNN) and findings claimed that the efficiently integrated by Raspberry Pi increased the robustness of detection and recognition [68]. According to findings of Table 9, the most share of studies developed by ML-based techniques for handling COVID-19 based dataset provided performance criteria. The most share of the performance criteria, according to Figure 8 is related to accuracy factor. Accuracy factor is a general and normalized factor. Therefore, it can be employed for comparing the ML-based methods with different datasets. Figure 9 presents the accuracy values for each model for comparing their performance in handling COVID-19 dataset.   Figure 9 indicates CNN with SVM classifier, Genetic CNN and pre-trained CNN followed by ResNet, provided highest accuracy values. On the other hand, the lowest accuracy was related to single CNN followed by XGboost and KNN techniques.

Challenges and limitations
Nowadays, when the world is struggling with COVID-19 disease, every innovation and technology is used to fight this disease. Like many other areas, healthcare requires the support of new technologies such as IoT, and ML. Exploring of disease-related dataset, data preparation, prevention, and control of infectious diseases, has become one of the main purposes of AI. IoT and ML have a vital personality in better understanding, dealing with the COVID-19 crisis, and discovering the COVID-19 vaccine. ML-based technology allows computers to predict the pattern and speed of disease transmission with the intelligence they have and by mimicking large amounts of data. This result-oriented technology is employed for proper screening, analysis, forecasting and tracking of current and potential future patients. AI uses information from people with coronary heart disease, improved and dead people as tracking data.
To combat the spread of the corona virus, IoT-based methods of communicating with patients provide transparency and a better understanding of how the virus is spread, and strengthen the treatment and research process. ML is one of the new technologies in tracking the spread of the virus and finding effective parameters in it. The ML method can successfully identify high-risk patients and predict the necessary measures to deal with possible infections to reduce the point of effect of the disease. MLbased methods can estimate the risk of patient mortality through previous analysis. This technique improves the planning, treatment, and reduction of patients and is a complementary medical tool that works with data and evidence. On the other hand, this technology improves decision making and reduces the cost of treatment and diagnosis. At the same time, in the field of medical imaging, ML tools helps to recognize the patterns in the images and strengthen the ability of radiologists to diagnose the possibility of disease and early diagnosis of the disease.
One of the main limitations of IoT, and ML-based techniques for applications in COVID-19 is the lack of a complete dataset. This can be due to the unique development of models by limited data for a specific application within the same data field. The purpose of using IoT, AI, or ML-based techniques is to solve a specific problem in the real world with a real application that requires the use of special hardware and equipment. There are limitations in the cost and availability of developing and equipping communication hardware in therapeutic, diagnostic, estimation, and forecasting applications for IoT technology or ML-based techniques. Also, there are limited best practices available for IoT developers. The lack of IoT edge authentication and licensing standards has led to restrictions on the application and enactment of laws, regulations, and policies in the use of this technology, and this has led to the absence of IoT-based incident response activities as the best methods. All of these limitations mean that there is still no focus on identifying ways to gain situational awareness of the security of IoT assets in a medical complex.

Conclusion
The present study categorizes the applications of IoT, IoT-ML, and ML-based techniques to tackle COVID-19-related problems. The main applications are categorized into monitoring, detection, identification, classification and diagnosis. Studying, comparing and investigating of these applications requires a proper judgment about the performance and effectivity of outputs. According to a deep consideration about the evaluation criteria, it has been investigated that the accuracy, followed by recall and precision parameters have owned the highest portion of the evaluation criteria employment for analyzing COVID-19 based dataset using IoT and ML-based techniques.
Most of the studies lack of numerical analysis for the method performance. One of the main reasons that can be the nature of IoT technique which goes through a practical process and shows its performance in practical applications. In all the applications, IoT could successfully cope with the tasks. Such that, IoT provided a fast and efficient approach of tracking the disease spread. Most of the studies developed by ML-based techniques for handling COVID-19 based dataset provided performance criteria. The most popular performance criteria is related to accuracy factor. It can be employed for comparing the ML-based methods with different datasets. According to the results, CNN with SVM classifier, Genetic CNN and pre-trained CNN followed by ResNet, provided highest accuracy values.
Policy-making in the field of COVID-19 disease to examine the weaknesses and strengths and vulnerabilities of society in terms of the penetration of pathogenic viruses can be considered as additional measures and future studies. On the other hand, the study of collective behaviors can also be considered as perspectives to complete studies in order to prevent similar social harms and reduce costs incurred and not to surprise human life.