Internet of Things‐based healthcare system on patient demographic data in Health 4.0

Fargana J.Abdullayeva

Institute of Information Technology,Azerbaijan National Academy of Sciences,Baku,Azerbaijan

Abstract Recently,the spread of COVID‐19 virus infection and the increase of people number with chronic diseases have attracted great attention all over the world.The detection and control of such diseases based on patient demographic data are considered to be a major problem.The key issue in the solution to these problems is the development of methods and algorithms to forecast wellness and categorise patients according to their healthy and unhealthy states.In this paper,a comprehensive analysis of machine learning approaches in the field of diagnosing COVID‐19 has been conducted,and for the detection of chronic diseases in patients,to identify symptoms of COVID‐19 virus infection in advance,and control the situation a healthcare system has been proposed.The constructed system provides real‐time monitoring of chronic diseases and COVID‐19 virus infection in patients.The proposed system consists of five layers:IoT sensor layer,Data transmission layer,Fog layer,Cloud layer,the Application layer.The system architecture in the Fog layer uses machine learning and deep learning algorithms to diagnose patients'diseases,to generate and send diagnostic and emergency alerts to users.The classification module of the system's Fog layer categorises the patient's health status into healthy and unhealthy classes.In this module,to classify medical data the Decision Tree,Random Forest,SVM,Gradient Boosting,Logistic Regression algorithms are used.The COVID‐19 dataset is used to test the effectiveness of the methods.The best results from the comparative analysis of the methods are obtained from the Decision Tree,Random Forest,and Gradient Boosting algorithms,which are recognised data points with high accuracy and on the accuracy metric reached 1.0,0.99,1.0 values,respectively.The classification of the other two SVM and Logistic Regression algorithms provided the worst results,and the accuracy score of both classifiers obtained a 0.89 value.

1|INTRODUCTION

On 31 December 2019,27 cases of pneumonia of unknown etiology were detected in Wuhan city,Hubei Province of China[1].World Health Organization named this epidemic disease COVID‐19 and registered it as one of the types of coronavirus.This viral infection is also known as SARS‐CoV‐2.The coronavirus is named SARS‐CoV‐2 by the International Committee on Taxonomy of Viruses.

COVID‐19 is a disease close to Severe Acute Respiratory Syndrome(SARS)and the Middle East Respiratory Syndrome(MERS)and affects the lower respiratory tract and appears in humans as pneumonia.Despite serious global quarantine measures,COVID‐19 continues to increase.

According to WHO statistics on 3 August 2022,coronavirus infection was registered in 216 countries around the world[2].Globally,from coronavirus,more than 577,018,226 people were infected,and 6,401,046 people died in the world[2].The spread of COVID‐19 infection in Azerbaijan started on February 28,2020[3].This date was registered by the operational headquarters of the Cabinet of Ministers of Azerbaijan as the first case of COVID‐19 coronavirus infection in the country.

In the Azerbaijan Republic,on August 3,2022,a total of 799,471 infections and 9752 deaths were registered by WHO[4].The dynamics of the disease spreading in the Azerbaijan Republic continue to grow.

On January 30,2020,WHO declared the COVID‐19 epidemic in China a state of emergency that could make a significant risk to countries with weak health systems.The Emergency Committee noted that the prevention of the spread of COVID‐19 infection could be achieved by the development of reliable early detection,isolation,emergency treatment,and a patient contact tracing system[5].

In Ref.[6],for the prevention and management of SARS‐CoV‐2 and COVID‐19 diseases 9 most important research issues have been identified.The development of diagnostic rules for the SARS‐CoV‐2 virus and methods for early detection of infected patients took place in the fifth position among the problems identified.

Vital signs of the disease in patients include dry cough,shortness of breath,fever,sore throat,and bilateral accumulation of fluid in the lungs.54.3% of patients infected with SARS‐CoV‐2 are men with an average age of 56.Complications are mainly observed in patients with an average age of over 56 and with chronic diseases such as cardiovascular,cerebrovascular,endocrine,digestive and respiratory.In patients in the intensive care unit the shortness of breath,dizziness,abdominal pain,anorexia cases are observed[7].Currently,no effective antiviral treatment or vaccine against COVID‐19 exists.

Recently,the number of people with chronic diseases is rapidly growing.Today,chronic diseases such as diabetes,heart disease,and hypertension are among the world's most actual economic and social problems,causing serious complications and even death in patients infected with COVID‐19.According to the World Health Organization,4.9 million people worldwide die from lung cancer,2.6 million from overweight,4.4 million from high cholesterol,7.1 million from high blood pressure,and 17.9 million from cardiovascular disease[8].Here the number of deaths from chronic diseases in the next 10 years is forecasted to increase 17%and to reach 64 million people.Patients with chronic COVID‐19 infection vary greatly from each other in their symptoms,prevalence,and treatment.

WHO and the US Centers for Disease Control and Prevention(CDC)have published recommendations for preventing the spread of COVID‐19 infection.However,these recommendations are of physical nature and include actions such as frequent hand washing,the usage of face masks,and not visiting risky places[9].

It is necessary to note that the mechanisms to combat the COVID‐19 coronavirus pandemic in China need to base on Artificial Intelligence,Big Data,and Internet of Things technologies,which are a manifestation of the fourth industrial revolution named Industry 4.0.In the approaches developed for the detection and control of the COVID‐19 coronavirus infection,considering machine learning and deep learning solutions is necessary[10].

These systems must remotely check the patient's symptoms based on data collected by sensors and make decisions,diagnoses,and prognoses regarding a person's infection with COVID‐19.These systems should contain components that can detect a person's chronic diseases early.

The prevalence of COVID‐19 virus infection made its early diagnosis a serious problem for healthcare organisations.One of the important steps in the management of COVID‐19 infection is the early detection of the disease and the timely implementation of preventive measures against these diseases.

We need to establish systems for remote detection and monitoring of COVID‐19 virus infection,including chronic diseases,to ensure real‐time control of the disease.The traditional healthcare systems cannot effectively control these types of diseases.

Currently,fewer studies have been done to determine the pathophysiological characteristics of COVID‐19[11],and there exists uncertainty about disease transmission ways.But numerous approaches have been proposed for the detection of chronic diseases[12–14].However,the proposed approaches have some limitations,such as not involving people in the system in the process of submitting clinical signs related to their health,not studying chronic diseases along with infectious diseases,not sharing the data recorded by medical institutions and the data recorded by people through various sensors,not considering the social media component that can enable patients to share their medical experiences,and the usage of traditional methods in the diagnosis of diseases.

In traditional methods of diagnosing diseases,patients have to go to the hospital for regular check‐ups,and doctors have to check every patient,regularly.In the period of disease epidemics spreading,meeting the needs of the emerging patient flow and providing them with appropriate medical staff is considered impossible.These problems require the development of remote monitoring systems for the provision of medical services.

The recent emergence of Mobile Computing,Cloud Computing,and Wearable Devices,which are a manifestation of the IV industrial revolution,has made it possible to design a number of cloud‐based healthcare services.Though most cloud‐based healthcare applications can provide real‐time remote services,these systems also face serious problems related to delays and Big Data transfers.There exists a high probability of errors during transmission as the data transmission is carried out over the network in these systems.Minor errors in data transmission result in inaccurate diagnostics and delays in notifying the patient.In the proposed approach,a Fog Computing layer is added between the Cloud layer and a user to overcome these problems.

The first layer of architecture consists of IoT‐based sensors.This layer implements data generation.The Fog computing layer provides a rapid analysis of data collected by IoT devices,real‐time decision making without any delays,prediction of COVID‐19 virus infection,detection of chronic diseases,and generation of real‐time alerts to patients.The Cloud layer provides storage and processing of data that cannot be processed in the Fog layer.

The proposed system aims to solve the problems faced by chronic disease detection systems and covers the issues related to the early detection of COVID‐19 virus infection.The scientific contributions of the work are as follows:

·For remote diagnosis of chronic diseases and COVID‐19 virus infection according to the patient's vital signs,a fog‐based system is developed.

·The architecture of the medical data classification module of the proposed healthcare system is developed.

In Section 2 problems of the traditional healthcare sector are described.Section 3 describes the related work.Section 4 introduces vital signs for patient monitoring.Section 5 describes the foundations of wearable devices and wearable physiological body sensors.Section 6 describes the concept of the Healthcare IoT ecosystem.Section 7 gives the architecture of the proposed method and describes the functions of its various components.In Section 8 the results of the experiments are provided.The conclusion of the paper is given in Section 9.

2|TRADITIONAL HEALTHCARE PROBLEMS

The traditional hospital‐centred healthcare system has several flaws:

‐Limited time.The growing number of patients and people with disabilities restricts the ability of doctors to examine the health status of each patient effectively.The short examination time of doctors isolates them from information about the patient's everyday actions such as body movement,diet,dreams,and public lifetime.These features are all considered significant in the diagnosis and treatment operation.

‐Adherence monitoring.Physicians are unaware of patients'adherence to the prescribed treatment,such as medications,rehabilitation sessions,and diet.Non‐adherence to treatment increases the patient's risk of hospitalization and this results in economic damage to the patient's family.

‐Ageing population.Compared to 2013,in 2050,a double increase is forecasted in the number of older(age over 60)people in the world.So,if this increase constituted more than 841 million people in 2013,this number is expected to exceed 2 billion in 2050[15].In such a situation,there is no doubt that the elderly population will require more resources for the provision of medical services.

‐Urbanisation.In 2015,WHO predicted that 70 percent of people of the worldwide will live in the town.To serve the growing population in big cities,there will be a requirement of lot of healthcare infrastructures.Here,also note that the big cities will become epidemic centres of infectious diseases that can easily spread among the dense population.

‐Lack of healthcare workers.The growing demand for medical services has also increased the demand for medical professionals such as doctors,surgeons,dentists,nurses,nurse assistants,professional nurses,and laboratory workers.The development of the remote healthcare infrastructure is considered an alternative solution to this challenge.

‐The increase in the cost of medical services.The increase in healthcare costs is the significant criteria in today's health industry.

3|RELATED WORK

Numerous studies are devoted to the remote monitoring issues of patients.In Ref.[16],to monitor the health status of the elderly and the people with disabilities,a system called the Healthcare Industrial IoT(HealthIIoT)is proposed.Health-IIoT is a combination of communication technologies,interconnected applications,people,and objects(devices and sensors)that operate together as a single smart system to monitor,track,and store medical data.The HealthIIoT monitoring system collects ECG‐type medical data via mobile devices and sensors and transmits it securely into the cloud for sharing among healthcare workers.To prevent identity theft and clinical errors by healthcare workers,the proposed methods such as signal amplification and watermarking are used.The efficiency of the proposed approach is tested by both experimental evaluations and the IoT‐based ECG‐type cloud monitoring service simulation.In the study,the authentication of the medical monitoring signals was carried out to ensure security.The main research object of this study is electrocardiogram monitoring.In Ref.[17],based on web services and cloud computing,a remote patient monitoring system is developed.

In Ref.[12],an Internet of Things‐based architecture of a patient's health status monitoring system has been proposed.The data collection,analysis,and visualisation are the key building blocks of the architecture.The proposed system uses wearable sensors to measure various physiological parameters called blood pressure and body temperature.The sensors transmit the collected information to the server via Bluetooth.This system stores the data on a remote server to allow healthcare professionals to later access the data over the Internet.Experiments were conducted on the monitoring of cardiovascular diseases.

In Ref.[13],a remote health monitoring system via the Web is developed.A C45 decision tree algorithm with a high confidentiality factor is developed for automatic decision making.The architecture contains a component that sends an automatic SMS alert to the patient's doctor if the patient's physiological symptoms are critical.The decision to send a warning signal depends on the outcome of the classification of sensor data.For the proposed method,training data contains parameters such as heart rate,respiration rate,oxygen saturation,blood pressure,ECG,and accelerometer.In this study,a constructed decision tree has been simulated on the WEKA software package,and it classifies the data into normal and anomaly classes.

Saha et al.[14]propose a system for monitoring the patient's health using IoT and cloud computing.The patient's pulse,heart rate,blood pressure,and temperature are selected as the main health indicators.The system mechanically generates a warning when it detects abrupt changes in a patient's health and sends it to the patient.To prevent fraud cases and clinical errors by doctors,by using the watermarking the improvement of the signals is provided.The proposed system consists of four layers:device layer,network layer,middleware layer,application layer.Diseases such as obesity,hypertension,arrhythmias,fever and diabetes can be detected through the established IoT system.As diabetes is diagnosed by analysing blood glucose,this process requires the patient to inject a physical needle.The system consists of three medical sensors:temperature sensor,pulse sensor and blood sensor.The user wears all sensors to collect the data.The collected data is transmitted to the cloud for processing.The result is loaded on a monitor connected to the system.BMI is calculated using the data.When a patient is diagnosed with a health condition,the system informs the user what disease may occur and what confrontational measures to take.

Kumar et al.[18]offer a cloud and IoT‐based mobile healthcare system for monitoring and diagnosing diabetes.In this study,diabetes is targeted and an algorithm for the Neural Network Classifier based on Fuzzy Rules is proposed to diagnose the severity of this disease.Large amounts of various data,such as images and text,are collected using IoT devices.This data is stored securely in the cloud and is accessed through healthcare applications.

Farahani et al.[19]proposed a conceptual model of Fog computing‐based IoT ecosystem(architecture)of E‐health.Three layers are included in this system.The study has proposed a method of detecting anomalies at an early stage to determine the worsening of the patient's health state.Detection of anomalies in the system is based on the algorithm of Hierarchical Temporal Memory(HTM).This method can detect temporal abnormalities,such as heart attack,stroke etc.

Georgaka et al.and Anzanpour et al.[20,21]developed the Early Warning Score System(EWS),which detects the deterioration of patient's condition prematurely.The goal of EWS is to analyse six clinical features.The evaluated clinical features are transformed into the composite risk value of the patient's deterioration.A combination of all scores allows for obtaining a composite index that reflects the overall risk of deterioration in patients.EWS is a globally accepted approach and is used in hospitals around the world.

Gómez et al.[22]proposed an ontology‐based Internet of Things architecture that monitors patient health and provides the necessary recommendations to patients with chronic disease.Here,the patient's health is monitored for three symptoms—diabetes,high blood pressure,and cardiac arrhythmia.

Doorsamy et al.[23]investigate the general architecture of IoT‐based healthcare systems,components,and subsystems of IoT systems in healthcare,and identifies technical issues related to IoT systems for smart healthcare.Opportunities are explored,and possible solutions are proposed to overcome existing problems.The proposed IoT‐based healthcare system consists of 3 main layers:the sensor layer,the middleware layer,and the server layer.

The aim of Ref.[24]is to develop IoT architecture for the Remote Monitoring System for elderly patients.The constructed system provides timely and accurate information on medical warnings,hospital visits,medications,and social support.In this system,data collected about the patient from the various remote sensors are combined,analysed,and converted into a data format that can be understood by the relevant persons(e.g.doctors and nurses)for immediate response and action in accordance with standard operating procedures.

Dang et al.[25]analyse the state of the art of healthcare system development,examines their security issues,discusses the problems and opens integration issues of IoT and cloud computing in healthcare.The study also proposes a new IoT approach for healthcare.In Ref.[26],a cloud‐based mobile health monitoring system has been proposed.Celesti et al.[27]proposes the Tele‐Rehabilitation as a Service(TRaaS)model.

Al‐Turjman et al.[28]examine the current technical and architectural problems of the healthcare industry,proposes an architecture of the Internet of Medical Things(IoMT)containing three‐component such as sensors,communication gateways(Zigbee,LTE Wi‐Fi),and the cloud.

Based on the convergence of the Internet of Things and cloud computing,Ref.[29]proposed a hybrid HRMM(Hybrid Real‐time Remote Monitoring)system for monitoring patient health status.For storing,processing and analysing big data,and constructing a classification model for the categorisation of a patient's health state,an HRMM uses the power of cloud computing.In the case of Internet interruption and cloud shutdown,the local block of the HRMM is used to predict the health status of the patient.The local block of HRMM uses classification models(such as SVM and NB)to solve this problem.In the study,the minimum feature selection is provided to obtain a high classification result by using the NB‐WOA hybrid model,which is a combination of the Naïve Bayes(NB)and Whale Optimisation Algorithm(WOA)algorithms.The termination of the HRMM monitoring process is ensured through the application of the NB‐WOA model.The proposed method is used to predict the health status of patients suffering from high blood pressure.The PhysioNet MIMIC‐II database is used to evaluate the effectiveness of the proposed approach.

Qi et al.[30]provide a systematic review of IoT‐based medical systems,identifies areas for research in these systems,and describes key technologies.Future directions and research problems in the field of IoT‐based medical systems are described.The four‐layer SOA(Service Oriented Architecture)architecture of an IoT‐based healthcare system is proposed:a sensor layer,a network layer,a data processing layer,and an application layer.Healthcare systems are classified in terms of a 4‐layered IoT system,and research problems are identified separately for each of them.Based on the proposed IoT architecture,the main methods used in sensor,network,data processing,and application layers are analysed.

Rodrigues et al.[31]examine the current state of research in the field of the Internet of Health Things(IoHT),describes the solutions created in this segment and identifies problems.Here,IoHT is divided into four categories:(1)remote health monitoring,(2)smartphone‐based health solutions,(3)ambient assisted living and(4)wearable devices.

Aceto et al.[32]investigate the current state of cloud application,IoT,wireless networks,Big Data,robotics,social networks,and 3D print‐based ICT technologies in healthcare.In this study,the ICT‐based healthcare paradigms include E‐health,mobile health,personalised health,smart health,health from anywhere,and comprehensive health.Types of medical data are described.

In Ref.[33],an IoT‐based Fog‐oriented cloud model for healthcare is proposed.The proposed system effectively manages the patient's data with heart disease and diagnoses the health state to identify heart disease.The model consists of the Body Area Sensor Network layer,IoT Devices layer,Fog Server layer,Resource Manager layer,and Cloud Data Center layer.The effectiveness of the proposed method is tested on iFogSim software of the CloudSim environment based on network engagement time,power consumption,and latency parameters.

Costa et al.[34]propose a conceptual model of a patient‐centred system that monitors a patient's health condition based on the Internet of Things.The system consists of four different layers:(1)collection,(2)storage,(3)processing,and(4)description.The basic component of this architecture is machine learning methods that perform the correlation of data and their conversion into useful information.Different approaches used in hospitals to collect and combine clinical features have been analysed.

Recently,the researchers,to advance the predictive analytical models differentiating COVID‐19 symptoms,are trying to develop artificial intelligence methods that can use texts on electronic medical records(EMRs)[35].

These methods can convert text type data taken from EMRs into symptom data.Such worldwide progress in the COVID‐19 diagnosis field necessitates the development of more effective methods for the classification of texts in disease diagnostics.One of the main directions of text analysis is keyword extraction.To represent text in a condensed way,Ref.[36]proposes a keyword extraction method.Keyword extraction enables the representation of the documents in compact form.Creating a compact representation of the text allows us to get a summary of the large‐scale text.Here,from the content of the text,the keywords are extracted,and they are used as features.Onan[37]proposes a deep learning approach that can divide features into multiple classes according to the importance degree of the features.The approach is based on bidirectional convolutional recurrent neural network architecture.The main idea here is enhancing important features in each group,while weakening the less important ones.For achieving better predictive performance,Ref.[38]proposes a consensus clustering‐based undersampling scheme.In this scheme,the number of instances in the majority class is undersampled by using a scheme based on consensus clustering.Onan[39]proposes an approach for the topic modelling in text analysis.This approach enables the identification of important topics within the text collection.The proposed approach combines the functions of word embedding schemes and cluster analysis.To implement the intellectual analysis of the texts,in Ref.[40],several separate feature lists were obtained by applying the different feature selection methods to the data.Then,by using a genetic algorithm,the aggregation of the feature lists was carried out.Onan et al.[41]propose an ensemble learning approach to build a more sustainable classification model for text classification.The ensemble pruning method was used to build the ensemble of classifiers.Onan[42]proposes a deep learning approach for the implementation of decision making from text‐based information collected from social networks.The architecture of the proposed model consists of TF‐IDF weighted Glove word embedding and CNN‐LSTM models.Numerous classification approaches based on topic modelling constructed on swarm intelligence optimisation have been proposed by Onan[43]for categorising biomedical texts.Here,the Latent Dirichlet allocation(LDA)method eliminates the problem faced by the large volume of the vector space model,while the optimal estimation of LDA's parameters and of the number of topics is performed by swarm intelligence.

Chadaga et al.[44]propose a machine learning approach that identified COVID‐19 disease based on analysis of blood platelets,leucocytes,monocytes,and eosinophil parameters.Four different classifiers are used to perform the classification;the Synthetic Minority Oversampling Technique(SMOTE)is used to generate synthetic samples,and the Shapley Additive Explanations(SHAP)method is used to calculate the weight of each feature.

Dash et al.[45]develop an intelligent computational model that predicts the spread of COVID‐19.The model,entitled as Facebook Prophet,can determine the dynamics of the spread of COVID‐19 over the next 90 days.

Rahman et al.[46]propose an architectural approach for the intelligent and effective management of smart industrial enterprises based on the IoT network under the COVID‐19 conditions.The architecture includes an SDN layer(such as data,control,and application)to effectively and automatically monitor the data of remote IoT devices.Here,the convergence between SDN and Network Function Virtualisation(NFV)provides an effective management mechanism for managing IoT sensor data.In architecture,sustainable integration of data is ensured.The architecture has the ability to reliably integrate data from devices required for Industry 4.0 in the context of the COVID‐19 pandemic.

Table 1 describes the scientific innovations of articles examining various aspects of healthcare systems.

4|VITAL SIGNS FOR PATIENT MONITORING

Since the early 20th century,healthcare workers in hospitals monitor patients'health state by measuring their vital attributes[47].These vital attributes include blood pressure,temperature,heart rate,respiratory rate,and oxygen saturation[48].The guideline of the National Institute for Health and Clinical Excellence(NICE)recommended monitoring a minimum of 6 vital signs:blood pressure,temperature,heart rate,respiratory rate,and oxygen saturation.This guideline recommends that in special cases additional parameters are also can be considered,such as the amount of urine excreted,the degree of pain,or other biochemical analysis[49].

TABLE 1 Comparative analysis of the related studies

In some studies,additionally to the main vital signs,parameters such as degree of pain,level of consciousness,and the amount of urine excretion are also considered[48].Besides,in Ref.[34]eight vital signs and their normal threshold values for monitoring patients'health status in hospitals are identified.

5|WEARABLE DEVICES

Wearable devices are smart devices that can attach to the body,for example,watches,shoes and body sensors.The types of wearable technologies are described in Figure 1[31].

To display signals,such as the patient's body temperature,heart rate and blood pressure these devices should have the ability to connect to the physiological sensors.

5.1|Wearable physiological body sensors

The characteristics of sensors that record the patient's vital signs and transmit them directly to the network or mobile device and their accepted threshold values for normal and anomalous conditions are described in Ref.[13].These sensors include blood pressure,body temperature,heart rate,respiration rate,oxygen saturation in blood(spo2)and ECG.

FIGURE 1 Different types of wearable technology[31]

Other sensors that record the patient's vital signs include[31]:

Diabetes.Glucose sensors,contextual sensors and near‐infrared led sensors are used to diagnose diabetes.

Asthma.Pulse and temperature sensors are used for asthma monitoring.

Heart diseases.Optical heart rate sensors,BP sensors and ECG sensors are used for heart diseases.

Hyperthermia and hypothermia.Thermopile Infrared(IR)sensors and wearable thermometry are used to diagnose hyperthermia and hypothermia.

Tele‐surgery.In telesurgery,microelectromechanical sensors(Mems sensors),robot arms,and microcontrollers are used.

Ebola.Lightweight body sensors and RFIDs are used to diagnose Ebola.

Wheelchair management.Wheelchair management uses a camera sensor,accelerometer sensor and power sensor.

COVID‐19 coronavirus infection.Currently,there does not exist a single sensor to detect COVID‐19 coronavirus infection.However,for the detection of this virus infection,sensors that are capable to record fever,cough,shortness of breath,persistent pain,chest pressure,bluish lips,and bluish face symptoms can be used.

Environmental.Water quality detector sensors,climate sensors,air temperature,humidity,carbon dioxide and mosquito density sensors are used to collect environmental data.

Drug.RFID tags are used to collect drug‐related information(strength,type,shape,and proportion).

Location data.GPS sensors are used to collect location data(location of mosquito dense sites,mosquito breeding sites,and time).

Meteorological data.A climate detector sensor is used to collect meteorological data(maximum temperature,minimum temperature,rainfall,and humidity).

In this study,we used the SARS‐CoV‐2 testing device and an X‐Ray image sensor to collect data.A total of 111 variables were used for training the algorithms[50]:age,gender,haematocrit,haemoglobin,platelets,red blood cells,lymphocytes,mean corpuscular haemoglobin concentration(MCHC),leucocytes,mean corpuscular haemoglobin(MCH),basophils,red cell distribution width(RDW),eosinophils,mean corpuscular volume(MCV),lymphocytes,monocytes,c‐reactive protein(CRP)etc.

6|HEALTHCARE IoT ECOSYSTEM

Medical data,location data,drug data,environmental data and meteorological data can be used to check patients'wellness.This type of data is collected through a variety of wearable devices and sensors and forms a patient's IoT‐based healthcare ecosystem.The components of the IoT‐based healthcare ecosystem are illustrated in Figure 2.

Here,we see stakeholders such as patients equipped with IoT sensors,medical professionals,healthcare research centres,social media,ambulance services,pharmacies,smart healthcare devices and house protection services by generating signals to collect data about the patient.

7|THE PROPOSED ARCHITECTURE

By conducting a comprehensive analysis of existing architectures,the general architecture of the monitoring system,which detects the presence of chronic diseases and COVID‐19 virus infection in patients in real‐time,is proposed as follows(Figure 3).

The system allows monitoring of patients'health conditions at home or in a healthcare institution.The proposed system consists of 5 layers:IoT sensor layer,Data transmission layer,Fog layer,Cloud layer and Application layer.

Data flow in architecture is carried out in the following steps:

Step 1.The wearable IoT sensor layer in architecture collects data from the various medical sensors,location sensors,body sensors,environmental sensors and meteorological(synoptic)sensors.Depending on the problem statement type,wearable devices such as an electrocardiogram(ECG),blood pressure,body temperature,blood oxygen level,pulse rate,electromyography,motion measurement(accelerometer),galvanic skin response,respiration rate(airflow),electroencephalogram and electrodermal activity,and many other sensors can also be used.

Step 2.The collected data is sent to the Fog layer through the Data transmission layer for real‐time processing and diagnosis of the patient's health condition.Once the patient's health condition is diagnosed,an alert signal is generated in the Fog layer and sent to the patient's mobile phone for timely prevention.In this layer,the patient's health status is classified into healthy and unhealthy classes.Classification is an important decision‐making mechanism in providing various medical diagnoses.The classification component performs the initial diagnosis by dividing patients into healthy and unhealthy classes,using machine learning classifiers.There are two types of alerts sent to patients'mobile phones:healthy or unhealthy.

FIGURE 2 Patient IoT‐based healthcare ecosystem

FIGURE 3 Real‐time monitoring system for chronic diseases and COVID‐19 virus infection

The diagnosis process of the patient's health condition is represented in Figure 4.

The proposed framework consists of two blocks:selection of features and classification of the health status of the patients.First,features are selected from the COVID‐19 dataset;then the generated vector of features is transmitted to the classification module.The vector of features more similar to the same health condition scenario is categorised in this module.

Step 3.The results of this analysis are stored in the cloud layer of the architecture,and at the same time,if the patient is diagnosed with a viral infectious disease,to determine the extent of the epidemic,through the social media component of the Cloud layer the patient's Social Network is analysed,based on the patient's social media data.The Cloud layer also generates warning messages about infected areas and sends them to healthy people.Considering the component that generates warnings about infected areas in the architecture allows government agencies to take the necessary countermeasures for population health and to control virus infection in the infected areas.

Although a number of architectures have been proposed for the diagnosis of COVID‐19,there are no architectures in the literature that address the issues of the Application layer and the Data transmission layer.Taking these components into account in the approach eliminates problems with big data transmission.There exists a high probability of errors during transmission as the data transmission is carried out over the network in the existing systems.Minor errors in data transmission result in inaccurate diagnostics and delay in notifying the patient.In the proposed approach,a Fog Computing layer is added between the Cloud layer and a user to overcome these problems.

During COVID‐19 disease,various complications occur in the health state of patients.The existing systems are constructed so generally that they are not considered dynamic enough for patients with COVID‐19 infection.The advantage of the proposed architecture over existing architectures is that the integration of mobile devices,body sensors,cloud computing,and healthcare workers'computers allows patients to monitor their health status adaptively.In addition to these features,another advantage of the proposed architecture is that it provides Big Data analytics and the availability of cloud infrastructure allows real‐time monitoring of patient's disease status by healthcare professionals remotely.

7.1|Medical data classification module architecture in the proposed healthcare system

According to patients'vital signs,machine learning and deep learning methods are used to classify them into healthy and unhealthy classes.

While classification,once the patient's health condition is diagnosed,the Fog layer provides the patient with a diagnostic and emergency warning message.

The architecture of the Fog layer classification module is illustrated in Figure 5.

FIGURE 4 Proposed health condition monitoring framework

FIGURE 5 The architecture of the Fog layer classification module

The predictions and diagnoses of the patients'health conditions for healthcare professionals are provided through applying machine learning algorithms to patient data,collected through sensors or involved from medical institutions.The aim of integrating the data collected in medical institutions into the system is to increase the accuracy of disease diagnosis.In future studies,classification and clustering methods based on machine learning and deep learning methods will be developed based on medical imaging data collected by various sensors to detect the presence of chronic diseases and COVID‐19 virus infection in the patient.

7.2|Machine learning

Decision Tree,Random Forest,SVM,Gradient Boosting and Logistic Regression algorithms are used to classify medical data.

8|EXPERIMENTS

The COVID‐19 dataset is used to test the effectiveness of the methods[50].The COVID‐19 dataset consists of 111 features and 5644 rows.The disease state of the patients listed in the dataset is classified into two classes,Positive(1)and Negative(0).A total of 5644 patients are registered in the dataset,of which 5086 belong to the negative class and 558 to the positive.In the experiments,80% of the data is used for training and 20% of the data is used for test purposes.For the detection of COVID‐19 cases in patients,all 111 features are fed into the entry of the classifier as a vector.The features includes Hb(haemoglobin)saturation,haematocrit,haemoglobin,platelets,red blood cells,lymphocytes,mean corpuscular haemoglobin concentration(MCHC),leucocytes,mean corpuscular haemoglobin(MCH),basophils,red cell distribution width(RDW),eosinophils,mean corpuscular volume(MCV),lymphocytes,monocytes,c‐reactive protein(CRP)and patient demographic data,such as name,address,date of birth and national unique patient identifier number.

The results obtained from the application of Decision Tree,Random Forest,SVM,Gradient Boosting and Logistic Regression algorithms into the dataset are included in Table 2.

Decision Tree,Random Forest and Gradient Boosting achieved the best results among the algorithms included in Table 2.Thus,the values,obtained by these algorithms over each metric are approximated into 1.This means that the algorithms are able to divide the two classes accurately.The other two algorithms,SVM and Logistic Regression,almost cannot work well in this dataset.Thus,the values they obtained in the positive class are quite low.And this is not considered a good result.Each algorithm gives different results on separate databases.Due to the small number of samples from the positive class,SVM and Logistic Regression algorithms could not recognise samples from this class and showed low results.

To perform a comparative analysis of the methods the ROC values of the algorithms are calculated,and for each algorithm,the ROC curve is constructed and represented in Figure 6.In Figure 6,Decision Tree,Random Forest and Gradient Boosting are obtained the same values,and their ROC curves overlapped each other and approximated into 1.Compared to other algorithms,SVM and Logistic Regression algorithms performed poorly.In Figure 6,these algorithms are represented in red and green colours,and over the ROC,obtained 0.67 and 0.73 values,respectively.Approximating the ROC curves into the midline is not a good state.The closer the curves into 1,the better the models are considered.

To evaluate the robustness of the algorithms the programme is executed 10 times(Table 3).

A boxplot graphic based on the average value of results obtained on ROC values is created and represented in Figure 7.

As seen in Figure 7,Random Forest,Decision Tree,and Gradient Boosting,compared to the other algorithms,obtained better results.Because the deviation of these algorithms is very small,the boxplot graphic of these algorithms is very dense.

The experiments are also provided on some samples of the Chest X‐ray images dataset[51].Normal and COVID‐19 type Pneumonia are illustrated in Figure 8.The experiments are performed on 100 images taken from the main database.

TABLE 2 Classification results of the machine learning methods

FIGURE 6 Comparison of the methods

TABLE 3 ROC values by the number of launches

Here,first of all,to unify the Chest X‐ray images the size of all images is formatted into 162×128.Because Chest X‐ray images are grey,no other pre‐processing is performed on them.

In this study,the used Chest X‐ray images are grey‐scale images.The pixels of these images are the features used to detect COVID‐19 type pneumonia.The value between 0 and 255 is considered the value of the grey portion,where 0 represents black and 255 represents white colours.In the model,at first,the bicubic interpolation method is applied to extract the features from the images,then these features are fed into the input of the classifiers,and the data are classified into COVID‐19 and normal classes.Logistic Regression,Decision Tree Classifier,Gaussian Naive Bayes,KNN,SVM and Deep CNN(Convolutional Neural Network)algorithms and ensemble architecture are used to implement the data classification.

To provide the classification of the data into COVID‐19 and normal classes a comparative analysis of the classical methods such as Logistic Regression,Decision Tree,Gaussian Naive Bayes,KNN,SVM and CNN with constructed ensemble learning method is provided.The results of the experiments are given in Table 4.

Logistic Regression,Decision Tree,Gaussian Naive Bayes,KNN,Random Forrest algorithms are used as the base models to determine the presence or absence of COVID‐19 pneumonia in X‐ray images.During the experiments,the KNN algorithm obtained the highest results,and its accuracy metric reached to 0.77417 value.Besides this,the Logistic Regression and Gaussian Naive Bayes algorithms also obtained high results,and their accuracy metric value for classification of X‐ray images reached 0.72612 and 0.70161,respectively.The SVM and Decision Tree algorithms showed relatively poor results and over accuracy metric achieved 0.63415 and 0.64807 values,respectively.

FIGURE 7 Boxplot graphic of machine learning methods for patient disease detection

In this study,to improve the results of the weak classifiers,the Soft Voting Ensemble and Hard Voting Ensemble models are developed.The developed models are based on the voting ensemble strategy and combine the above mentioned five classifiers.As a result of the experimental study of these models on X‐ray images,the Soft Voting Ensemble model achieved a value of 0.76612 on the accuracy metric,and the Hard Voting Ensemble model achieved a value of 0.75806.Note that for the detection of COVID‐19 type pneumonia on X‐ray images the Deep CNN model is also investigated.Thus,the CNN model achieved 0.7467,0.8498,0.6632 and 0.7450 values in terms of accuracy,precision,recall and F1‐score,respectively.In Ref.[29],deep learning models such as LeNet5 CNN and LeNet5 CNN with Aug were tested on the dataset collected in this study.However,the results of these models were much weak from the results of the approach we proposed in this study.Thus,LeNet5 CNN,LeNet5 CNN with Aug and VGG16 models achieved 74.19,76.61 and 72.58 values,respectively,in terms of accuracy metrics.In our approach,the Soft Voting Ensemble method achieved a 0.76612 value,and the Hard Voting Ensemble method reached 0.75806.

The complexity analysis of the proposed method is provided based on runtime analysis.The number of iterations was taken 30.Here,41 s have been spent to the training of the model in every iteration.Since the total number of iterations is taken 30,overall 20 min were spent to the testing of the model.

TABLE 4 Results of the experiments for the diagnosis of COVID‐19 type pneumonia in the lungs

FIGURE 8 Normal chest X‐ray image(a)and image with COVID‐19 pneumonia(b)

9|CONCLUSION

In this paper,a remote monitoring system for the early detection of chronic diseases and COVID‐19 virus infection in patients is proposed.The system uses machine learning and deep learning methods to classify patients'conditions into healthy and unhealthy classes.For providing patient health status diagnosis Decision Trees,Random Forest,SVM,Gradient Boosting and Logistic Regression algorithms are used.The results of the experiments on the COVID‐19 dataset showed that the Decision Tree,Random Forest,and Gradient Boosting algorithms achieved the best results compared to SVM and Logistic Regression.The evaluation of the proposed approach on the Chest X‐ray images dataset is also provided.Logistic Regression,Decision Tree Classifier,Gaussian Naive Bayes,KNN,and SVM algorithms and ensemble architecture are used to implement the data classification.By applying these methods data are classified into COVID‐19 and normal classes.A comparative analysis of the classical methods such as Logistic Regression,Decision Tree,Gaussian Naive Bayes,KNN,and SVM with constructed ensemble learning method is provided.While applying methods to the diagnosis of COVID‐19 disease,compared to the classical classification algorithms the ensemble learning method developed in this study showed better results.Soft Voting Ensemble and Hard Voting Ensemble were used as the ensemble method.From the experiments of these ensemble methods applied to real datasets,superior results are obtained compared to separate algorithms.In future studies,it is expected to create a Graphic User Interface to help radiologists to diagnose COVID‐19.

ACKNOWLEDGEMENTS

This study was funded by the project“Management of big data resources in the electronic socio‐technological environment,the formation of electronic demography and the development of its intellectual analysis technologies”.

CONFLICT OF INTEREST

There is no competing interest related to the paper.

DATA AVAILABILITY STATEMENT

The data is available on the Internet.

ORCID

Fargana J.Abdullayevahttps://orcid.org/0000-0003-2288-6255