An intelligent framework for monitoring dengue fever risk using LDA-ANFIS

Abstract

Dengue fever (DENF) is considered as the world’s fastest-growing vector-borne disease. Its timely and proper treatment possesses great significance, as it affects vital body organs and causes morbidity and mortality. In this paper, a dew-cloud assisted DENF-based cyber physical system (CPS) is proposed. This system diagnoses DENF, and monitors the effects of DENF infection on vital body organs. This system uses Linear Discriminant Analysis-Adaptive Neuro Fuzzy Inference System (LDA-ANFIS) classification at the dew space to diagnose the DENF, and monitor the coronary heart disease (CHD) risk of the DENF infected users in real-time. Based on the real-time diagnosis and monitoring, the system immediately alerts the concerned stakeholders and provides timely medical support in case of any dengue-related emergency. The proposed CPS also assesses the Temporal Network Analysis (TNA)-based DENF outbreak risk, at the cloud space to control and prevent the outbreak of DENF by identifying the critical areas and individuals, which contributes in spreading DENF infection. The experimental evaluation of the proposed system acknowledges performance efficiency and utilization through various experimental and statistical approaches.

Keywords

Dew computing cloud computing coronary heart disease Linear Discriminant Analysis Internet of Things Adaptive Neuro Fuzzy Inference System

1. Introduction

Dengue fever (DENF) is considered as the world’s fastest-growing vector-borne disease. It is transmitted into humans by infected female mosquito Aedes agypti. A mosquito gets infected if it bites a person who is already suffering from dengue viral, and later this mosquito transmits the infection to healthy people by biting them. Within a week, an infected person can transmit the infection to others. DENF is significantly re-emerging disease in tropical and subtropical areas, especially during the rainy season. In the past few years, viral cases have increased in developing countries because of poor management of housing, water, and lack of vector detection, control, and prevention programs. The symptoms of DENF include fever, pain behind eyes, vomiting, severe headache, joint pain, soft bleeding, muscle pain, severe abdominal pain, nausea, and skin rash [34]. It is estimated that only 96 million out of 390 million dengue infection cases are clinically manifested every year, and around 3.9 billion people are at high risk of DENF infection [39]. Based on the severity of the DENF, it is necessary to control the infection rate by identifying possibly infected individuals and providing medical support to them at an early stage. The epidemic effects of dengue make real-time diagnosis an emerging issue for medical agencies in developing countries.

At present, healthcare systems are not that efficient in controlling dengue type infectious diseases. Present systems have various limitations like inadequate readily accessible diagnostics methods, hesitation to share health-related information and traditional methods for treating diseases. In addition to this, some users are incapable of visiting hospitals. Hence, to control this infection in real-time, remote recognition and monitoring are crucial for health service providers.

With the dramatic increase of DENF infected cases, many other complications are reported in the human body. Dengue affects various vital organs of the human body like liver, lungs, heart, and kidney [25]. Its effect on the heart is very critical, as any abnormal functioning of the heart may lead to death immediately, especially during coronary heart disease (CHD). The CHD develops when coronary arteries become too narrow, and blood flow to the heart reduces. In 2016, an estimated 17.9 million people died due to cardiovascular disease, out of which 7.4 million died due to CHD [7]. If patient history shows an increase in blood pressure, cholesterol, and heart rate along with DENF, then there are more chances of heart attack /stroke [38]. Henceforth, the monitoring of persons affected by CHD during their DENF period requires utmost attention in the domain of dengue-based healthcare systems. The timely diagnosis and medical treatment of CHD can cut down the death rate.

As the population is growing at a fast pace, it is hard for countable doctors to check and monitor each patient at fixed, regular intervals physically. It is often witnessed that the majority of the people falling with dengue may lose life within a short span of time due to the inability to reach the hospital on time. These limitations can be overcome by using Internet of Things (IoT)-based cyber physical systems. These systems enable the remote and real-time diagnose and monitoring of health attributes. Many researchers have proposed and developed smart healthcare systems like smart medical devices [33], mobile and remote medical care systems [23] etc. with the use of IoT devices, and cloud computing [17]. Though these systems provide multiple healthcare services in real-time, however, the huge quantity of data generated by IoT based systems face few difficulties to manage and process data stored at the cloud in real-time because of transmission delay and communication overhead [33]. Here, dew computing [28,31] is introduced in the proposed system for more flexibility, interoperability and usability. The major advantages of dew computing are that registered users are not only capable of storing and retrieving health-related information to and from the cloud, but also process data in real-time, near to the user [31], whether the internet connection is available or not. Processed data is synchronized to the associated cloud server once the internet connection is set up. The advantages of dew computing as compared to Fog computing are the higher computation capacity, lesser resource hunger, and greater decision capability. The majority of the cloud services are used to store collected data and information, which require internet facility all time. However, the internet facility may not be available all the time due to certain reasons (natural disasters, or certain problems at user end). Hence, the healthcare systems will be better used if IoT devices are assembled with dew computing.

Henceforth, in this paper, a dew-cloud assisted CPS is proposed for diagnosing DENF, and monitoring the CHD risk of DENF infected users in real-time. The proposed CPS uses dew computing in the cyber space for real-time determination of the health risk using LDA-ANFIS based classification and generates immediate alerts to the concerned stakeholders. The real-time computation of dew computing provides immediate alerts and on-time medical support in case of any dengue-related emergency. The proposed system also uses cloud computing for providing protection to users’ sensitive data against any unauthorized disclosure and assessing the geographical risk of DENF outbreak using temporal network analysis (TNA). The proposed system not only diagnoses the DENF infection but also monitors the working of vital organs (heart), during the infection period.

CPS is an emerging high potential computing paradigm that is gaining popularity in healthcare applications. It facilitates the interaction between the cyber space where information processing is done and the physical space where we all live and interact. Thats why the proposed CPS is focused on achieving the following objectives.

To develop a dew-cloud assisted DENF healthcare system that also considers the effects of DENF on vital organs of the human body.

To provide real-time remote diagnosis of DENF and monitoring of CHD patients during their DENF infection period.

To accurately collect, handle, and use CHD related parameters to identify the early risk of CHD during DENF infection.

To provide real-time diagnostics and warning alerts to concerned stakeholders for providing timely medical aid.

To provide data privacy to the sensitive information of the user.

To effectively assess the global DENF outbreak risk using the location of users and mosquito sites.

This paper is divided into five sections. Section 2 includes the related work of dengue fever, coronary heart disease, effects of dengue on the human body, and IoT based healthcare systems. This section also includes state of the art for technologies like cloud computing and dew computing. Section 3 discusses the proposed cyber physical system. Section 4 provides the experimental evaluation details of the presented paper, and discusses the obtained results. Section 5 concludes this paper.

2. Related work

This section includes some significant related works, related to four diverse directions of the proposed system.

2.1. Dengue fever

In 2013, Tunmibi et al. [35] designed a rule-based expert system for the diagnosis of fever. This model uses simple if/then rules and create a knowledge base of fever and then use Vb.net programming language to make new deductions. In 2014, Huang et al. [18] proposed an algorithm for distinguishing influenza-infected patients from dengue and fever-related diseases. In 2017, Thein et al. [34] discussed various case studies during the period from 01 January 2004 to 31 December 2008, of adult patients registered in hospitals of Singapore having dengue and also discussed its various symptoms. In 2015, Reddy and Kavitha [30] proposed an artificial neural network-based system to predict dengue at an early stage. Their model works in three steps. In the first step, health-related parameters are filled manually, then the related attributes are selected and finally make use of ANN to predict dengue infection. In 2016, Kumar et al. [37] analyzed dengue virus in metropolitan areas like Delhi, which helps in guiding and managing preventive measures against the dengue virus. In 2017, Rather et al. [27] discussed various methods to control and prevent dengue viral infection.

2.2. Coronary heart disease

In 2010, Karaolis et al. [20] provided ten major parameters for detecting the risk factors of CHD. In 2014, Yang et al. [42] proposed an optimized system based on ANFIS and LDA for CHD. In 2013, Xia et al. [40] discussed the electrocardiography (ECG) of the patient and used a model to find any abnormalities using ECG readings. In 2013, Chakraborty et al. [8] used RSSI in the proposed model to monitor the mobility of patients. In 2015, Lyu et al. [22] proposed CHD evaluation method based on linear time invariance. The following articles represent the effects of dengue on vital organs of the human body. In 2015, Samanta and Sharma [32] discussed the effects of dengue on the liver. In 2016, Virk et al. [38], analyzed, and concluded that with severe dengue, the cardiac-related complications must be monitored carefully to prevent mortality. In 2016, Arora and Patil [3] had concluded that if the patient is confirmed with dengue, then there are more chances of an increase in the cardio vascular related disturbance, which can cause death. In 2018, Kularatne et al. [21] concluded that patients with heart, liver, and dengue disease must be detected as well as monitored properly to prevent death.

2.3. IoT and cloud computing based healthcare CPS

This section presents the contributions of IoT and cloud-assisted cyber physical systems for healthcare. Healthcare cyber physical system plays a vital role in existing medical applications. In 2012, Bogdan et al. [5] designed a pacemaker controller to control heart rate variability using CPS. In 2014, Haque et al. [16] had discussed the cyber physical systems in health-care. In 2014, Xu et al. [41] proposed resource-based data acquisition and processing methodology for IoT data during emergency medical services. In 2015, Prittopaul et al. [26] explained about heart attack detection using ECG signals thorugh cyber physical system. In 2015, Giger et al. [13] proposed a health monitoring system for remote patients. In 2015, Ibrahim et al. [19] discussed the use of biomedical technologies to monitor various diseases like dengue. In 2016, Costanzo et al. [10] aimed at providing a reliable monitoring model based on embedded systems and mobile devices. etc. In 2016, Ahmad et al. [1] proposed a framework of Health-Fog for real-time analysis of health-related data. The proposed model remotely supervises different essential symptoms of the disease, and based upon the alarming parameters, it facilitates remote doctors to intervene. In 2017, Zhang et al. [44] proposed healthcare cyber physical system assisted by cloud and big data. In 2017, Andriopoulou et al. [2] proposed a IoT and Fog computing based model and illustrated its advantages. In 2018, Dey et al. [11] discussed the use of cyber physical system for medical purposes. In 2019, Paoletti et al. [24] developed an algorithm to prevent reprogramming attacks on implantable medical devices used to detect cardiac signals.

2.4. Dew computing

In 2016, Rindos and Wang [31] discussed the challenges and potentials of dew computing, and compared it with cloud computing. In 2017, Gusev [14] discussed in detail about the dew based architecture for solving IoT enabled applications. In 2018, Axak et al. [4] proposed dew-fog-cloud based architecture for providing health services. In 2017, Ray [28] proposed dew-cloud based architecture and also compared various technologies like fog computing, edge computing, cloud computing along with their objectives. In 2019, Ray et al. [29] explained the benefits of IoT and the dew computing-based system for real-time e-health care services.

3. Proposed system

The architecture of the proposed DENF monitoring CPS is shown in Fig. 1. It works in two spaces: physical and cyber. The physical space acquires person, health, and environmental data from handheld mobile devices, body sensors, and environmental sensors, and transmits it to the cyber space. The physical space also provides real-time and global-wide dengue-related diagnostic and warning alerts to users, government, and healthcare agencies. The cyber space houses various data processing-related modules at two subspaces: dew space and cloud space. The dew space employs dew computing for real-time DENF diagnosis and monitoring of CHD affected dengue patients. The dew space provides real-time diagnosis, monitoring, and medical support in case of any medical emergency. The cloud space employs cloud computing for sensitive information protection and TNA-based DENF risk outbreak assessment. The cloud space provides scalable global computation resources for area-wide processing of dengue-related data for providing privacy to user-related sensitive information, and identification of DENF affected and risk-prone areas. The data flow in the proposed CPS is shown in Fig. 2. Each space of the proposed CPS is discussed in detail as follows.

Fig. 1.

Architecture of DENF-based cyber physical system.

Fig. 2.

Data flow in the proposed CPS.

Table 1

Personal, health related and environmental factors

Attribute Type	Attribute Name	Acquisition Method	Sensors
Personal	UID	MobApp
	Name	MobApp
	Age	MobApp
	Sex	MobApp
	Workplaceś address and location	Sensor	GPS sensor
	Residentialś address and location	Sensor	GPS sensor
	Mobile Number	MobApp
	Occupation	MobApp
	Relatives	MobApp
Health Related	Fever	MobApp
	Severe Headache	MobApp
	Pain Behind Eyes	MobApp
	Nausea	MobApp
	Vomiting	MobApp
	Severe Abdominal Pain	MobApp
	Skin Rash	MobApp
	Joint Pain	MobApp
	Muscle Pain	MobApp
	Soft Bleeding	MobApp
	Systolic Blood Pressure	Sensor	iHealth BP7
	High Density Lipoprotein	Sensor	CardioCheck Plus
	Blood Glucose	Sensor	OEM Meter
	Age	MobApp
	ECG	Sensor	TAGecg
Environmental Related	Mosquito Density	Sensor	Mosquito
	Mosquito Dense Site	Sensor	Sensors
	Mosquito Breeding Site	Sensor
	Breeding Sites Count	Sensor
	Water Site Temperature	Sensor	Climate
	Water Site Humidity	Sensor	Sensors
	Water Site Carbon Dioxide	Sensor

3.1. Physical space

The responsibility of physical space is to acquire user data and environmental data. This layer acquires user data from handheld mobile devices, and body-attached health sensors of the users, whereas acquire environmental data from mosquito sensors, and other environmental sensors, as shown in Table 1. The user data includes the personal attributes, and the health data includes the DENF-related attributes and heart-related attributes of the user. The environmental data includes the attributes related to the mosquito sites and favorable conditions of mosquito breeding around the water sites. The handheld smart mobile devices provide an interface to communicate with both physical and cyber space via communication technologies such as GPS, Bluetooth, Wi-Fi, WLAN, and many others. These smart devices acquire the personal and health attributes of the users through the mobile application (MobApp) and body-attached health sensors from physical space and transmit it to the cyber space for health-oriented data processing. These devices also provide diagnostic and warning alerts from the cyber space located health-oriented analyses to the various stakeholders (users, relatives, government agencies, and healthcare service providers) present in the physical space.

This space acquires data in two ways: manual and automatic. For DENF-related attributes, space has to depend upon the users, whereas for heart-related and environmental attributes, the data is acquired automatically. The users initially register with the system. The systems grant a unique identifier (UID) to the users on their successful registration. This UID is used between various stakeholders, and modules, for any healthcare-related data processing and communication. Users provide personal information, and periodically the DENF-related information manually to the system through MobApp, whereas the heart-related attributes are acquired automatically from body-attached sensors through the handheld mobile device. The environmental sensors directly transmit the environmental attributes to the cloud space using IoT.

3.2. Cyber space

The cyber space houses various data processing-related modules at two subspaces: dew space and cloud space. The dew space employs dew computing for real-time DENF diagnosis and monitoring of CHD affected dengue patients. The dew space provides real-time diagnosis, monitoring, and medical support in case of any dengue-related medical emergency. The cloud space employs cloud computing for sensitive information protection and TNA-based DENF risk outbreak assessment. The cloud space provides scalable global computation resources for area-wide processing of dengue-related data for providing privacy to user-related sensitive information, and identification of DENF affected and risk-prone areas. Each subspace of the cyber space is explained as follows.

3.2.1. Dew space

This subspace of cyber space collects data from physical space, processes the ECG signals of the heart, and provides real-time health-oriented data processing to diagnose the incidence of DENF and CHD risk of DENF infected users. This space helps in notifying the concerned stakeholders regarding the health risk of the users and provides immediate medical support in case of dengue-related medical emergency. This subspace has four modules: data collection, ECG processor, LDA-ANFIS based health risk assessment, and alert generation. Each module of this subspace is explained in detail as follows.

3.2.1.1. Data collection An efficient and accurate healthcare system heavily relies on the accuracy of its modules to identify and analyze all causing factors. The proposed CPS has divided the causing factors into three categories, namely personal, environmental, and health factors, as specified in Table 1. The responsibility of the data collection module is to collect data regarding these three factors from the physical space. This module collects the data and forwards the ECG signals to the ECG processor to extract the useful insights for further processing.

3.2.1.2. ECG processor Some attributes are acquired from physical space in a form, which might require some pre-processing to extract usable insights. ECG signals are one such attribute that requires pre-processing to extract heart-related attributes like heart rate. Dew space provides this functionality through the ECG processor. This processor works in three phases: data validation, extraction, and classification, to extract heart-related attributes. The phased functionality of the ECG processor is explained as follows.

Data Validation

The ECG signals acquired from body attached sensors includes significant amount of external influences like noise and artifacts, which originated from both physiological and non-physiological sources. The physiological noise includes movement of body, shivering etc. captured during the acquisition of ECG signals. The non-physiological noise includes environmental noise, disturbance in electronic components, and line power. The ECG processor uses low pass and band filter for removing these noises from ECG signals.

Extraction

After validating the ECG signals, the ECG processor requires relevant features to detect any abnormality of the heart functioning. Henceforth, the extraction phase of the ECG processor extracts relevant features from ECG signals. ECG signals comprise various kinds of waves such as P, T, Q, R, and S wave. These wave’s intervals are used to diagnose the heart functioning. Among all, the four waves are most popular for heart functioning diagnosis i.e. RR, PR, QT, and QRS. ECG signals of the DENF infected users are acquired continuously, which results in large amount of data and takes more time for timely decision making.

The ECG processor in the dew space uses Fast Walsh Hadamard Transform (FasWHTrans) [12] to extract and highlight the discriminating features of ECG signals such as RR, PR, QT, and heartbeat rate. The FasWHTrans extract the features accurately in lesser time. It decomposes an ECG signal into a cluster of square waveforms or rectangular with values of either +1 or −1. This decomposition is used to measure frequencies from the original signal. This helps FasWHTrans to correctly identifies the sharp discontinuities in the ECG signals. The FasWHTrans consume lesser computation time taking fewer coefficients. The FasWHTrans in the proposed CPS is used to extract the number of pulses, and determine the heart rate using Eq. (1). $\begin{matrix} (1) & Heart Rate = \frac{60}{RR interval} \end{matrix}$ The FasWHTrans of a N-length ECG signal $e (t)$ is represented in Eq. (2). $\begin{matrix} (2) & F_{n} = \frac{1}{N} \sum_{j = 0}^{N - 1} e_{j} w (n, j) \end{matrix}$ where, $w (n, j)$ is a Walsh function.

To remove any probable errors that may occur because of inadequate extracted features, the extracted features from the ECG signals are normalized as represented in Eq. (3). $\begin{array}{l} η_{i} = \frac{F_{i} - m d}{s d}, \\ (3) & \forall F_{i}, i = 1, 2, 3, \dots, n \end{array}$ where $m d$ and $s d$ represents mean and standard deviation respectively.

Classification

Once the distinctive waves are extracted from ECG signals, an automatic classification of the extracted waves into the category of normal or abnormal state of the heart, is one of the distinctive feature of the proposed ECG processor. The ECG processor at the dew space uses unsupervised classification to analyze the data in real-time, where the classification has to rely on the information available in the ECG data, and there is no requirement of the prior knowledge. This system uses Gaussian process classification. It is based on the Laplace approximation. This choice was made because of its capability to handle large databases. In the Gaussian process, the clustering is used before the classification to formulate the training dataset. The diagnosis of heart status using the Gaussian process classifier [43] is represented in Eq. (4). $\begin{array}{l} P (F_{i} | G_{i}) \\ (4) & = exp (G_{i, c}) (\sum_{i = 1}^{c} exp (F_{i}^{T} G_{i}) - 1 \end{array}$ where $G_{i} = {[G_{i, 1} \dots G_{i, c}]}^{T}$ represents a latent function vector relating to ith data point, and $F_{i} = {[F_{i, 1} \dots F_{i, c}]}^{T}$ represents a corresponding target vector comprises correct class entry for ith observation, otherwise zero entries.

Fig. 3.

Layered structure of LDA-ANFIS.

Table 2

Heart-related fuzzy parameters

S. No	Heart Attributes	Range	Fuzzy sets
1	Systolic Blood Pressure (HA1)	${< 120, 120 – 150 ⩾ 150}$	{Normal, Mild, High}
2	High Density Lipoprotein (HA2)	${< 40, 40 – 59, ⩾ 60}$	{Normal, Mild, High}
3	Blood Glucose (HA3)	{<120, ⩾120}	{No, Yes}
4	Age (HA4)	{20–35, 36–50, >50}	{Young, Middle, Old}
5	Heart Rate (HA5)	{⩽100, >100}	{Normal, Abnormal}

3.2.1.3. LDA-ANFIS based health risk assessment This module utilizes LDA-ANFIS in dew space to classify the health risk of the registered users in real-time and provide immediate medical support in case of dengue-related medical emergency. Figure 3 shows LDA-ANFIS based classification architecture, in which DENF-related categorical attributes are fed into LDA for possible diagnose of DENF, and heart-related continuous attributes are fed to ANFIS for the diagnose of CHD risk of the DENF infected users. Based on the outcomes of both classifiers, the final health risk of the users is determined, as shown in Algorithm 1. This module classifies the health risk of the registered users as (I) No Risk (NoR), where the users in not infected from DENF, (II) Low Risk (LoR), where the user is infected from DENF, and his/her CHD risk is normal, (III) Medium Risk (MeR), where the user is infected from DENF and his/her CHD risk is mild, and (IV) High Risk (HiR), where the user is infected from DENF, and his/her CHD risk is high. The explanation of this module is presented as follows.

Linear Discriminant Analysis based DENF Classification

LDA is based on the idea of maximizing the ration of intra-low and within-class scatters, which is achieved by projecting the features to higher-dimensional space. The categorical variables that are given as input to LDA are Fever, Severe headache, Pain behind eyes, Nausea, Joint pain, Muscle pain, Soft bleeding, Vomiting, Severe abdominal pain, and Skin rash. Let GP1 and GP2 are two groups of size N1 and N2. Average vector $Y_{j}$ of the observed vector $Y_{j k}$ ( $j = 1, 2$ ; $k = 1, 2, \dots, N_{j}$ ), which fall in group ${GP}_{1}$ and ${GP}_{2}$ are determined using Eq. (5) and $X_{j}$ is covariance matrix which is evaluated using by Eq. (6). $\begin{array}{l} (5) & Y_{j} = \frac{1}{N_{j}} \sum_{k = 1}^{N_{j}} Y_{j k} \\ (6) & X_{j} = \frac{1}{N_{j} - 1} \sum_{k = 1}^{N_{j}} {(Y}_{j k} - {\overline{Y}}_{j}) \end{array}$

Linear transformation is applied to Y, which separates two groups as specified by Fisher discriminant method. $\begin{matrix} (7) & Z = d^{'} Y = d_{1} Y_{1} + d_{2} Y_{2} \dots d_{p} Y_{p} \end{matrix}$

Statistical value of text taking average of modified Z value of two group are determined as $\begin{array}{l} (8) & t = \frac{d^{'} (Y_{1} - Y_{2})}{{[d^{' x p} d (\frac{1}{n_{1}} + \frac{1}{n_{2}})]}^{\frac{1}{2}}} \\ (9) & X_{p} = \frac{(N_{1} - 1) Y_{1} + (N_{2} - 1) Y_{2}}{N_{1} + N_{2} - 2} \end{array}$

The maximization of $t^{2}$ is achieved by choosing canonical discriminant coefficient vector for above equation. Moreover, $d^{'} Y$ is determined using $\begin{matrix} (10) & d = X_{p}^{- 1} (\overline{Y_{1}} - \overline{Y_{2}}) \end{matrix}$

The maximum value of $t^{2}$ is represented as, $\begin{matrix} (11) & (\frac{1}{N_{1}} + \frac{1}{N_{2}}) {(\overline{Y_{1}} - \overline{Y_{2}})}^{' X_{p}^{- 1}} (\overline{Y_{1}} - \overline{Y_{2}}) \end{matrix}$

Assume that $d^{'} \overline{Y_{1}}$ and $d^{'} \overline{Y_{2}}$ are mean of transformed groups, $Y_{0}$ is classified as follows.

If $d^{'} Y_{0} > \frac{d^{'} \overline{Y_{1}} + d^{'} \overline{Y_{2}}}{2}$ , $Y_{0}$ is classified into group ${GP}_{1}$

If $d^{'} Y_{0} < \frac{d^{'} \overline{Y_{1}} + d^{'} \overline{Y_{2}}}{2}$ , $Y_{0}$ is classified into group ${GP}_{2}$

The LDA classifies the category of users as possibly DENF infected (DenF) if classification depicts

{GP}_{1}

, otherwise the user is diagnosed as an uninfected of DENF (N_DenF).

ANFIS-based CHD Risk Assessment

The structure of ANFIS is the amalgamation of fuzzy logic and artificial neural network. This adaptive network learns the parameter values for node functions employing training data. The input set of five-layered structure consists of Systolic Blood Pressure (HA1), High Density Lipoprotein (HA2), Blood Glucose (HA3), Age (HA4), Heart Rate (HA5) and one output. The output distinguishes the CHD risk of the registered user. Fuzzy sets and range of all the input variables are computed by the Trapezoidal membership function shown in Table 2. The mixture of least square and back propagation is used for learning. The explanation of the ANFIS for determining the CHD risk is as follows.

Layer 1: This layer consists of nodes that give membership of fuzzy set as follows. $\begin{matrix} O_{j}^{1} = μ_{A_{j}} (y) \end{matrix}$ where y is input to node j, $A_{j}$ denotes the associated fuzzy sets and $μ_{A_{j}}$ is the membership function. The considered scenario has five inputs that are fuzzified as,

$O_{j}^{1} = μ_{A_{j}} (HA 1)$ for $j = 1, 2, 3$ and $A_{1} = Normal$ , $A_{2} = Mild$ , $A_{3} = High$

$O_{j}^{1} = μ_{B_{j}} (HA 2)$ for $j = 1, 2, 3$ and $B_{1} = Normal$ , $B_{2} = Mild$ , $B_{3} = High$

$O_{j}^{1} = μ_{C_{j}} (HA 3)$ for $j = 1, 2$ and $C_{1} = Yes$ , $C_{2} = No$

$O_{j}^{1} = μ_{D_{j}} (HA 4)$ for $j = 1, 2, 3$ and $D_{1} = Young$ , $D_{2} = Middle$ , $D_{3} = Old$

$O_{j}^{1} = μ_{E_{j}} (HA 5)$ for $j = 1, 2$ and $E_{1} = Normal$ , $E_{2} = Abnormal$

Layer 2: This layer calculates the firing strength of the fuzzy rules as, $\begin{array}{rcl} O_{j}^{2} w_{j} & = & μ_{A_{j}} (HA 1) * μ_{B_{j}} (HA 2) \\ * μ_{C_{j}} (HA 3) * μ_{D_{j}} (HA 4) \\ * μ_{E_{j}} (HA 5) \end{array}$ where, $j = 1, 2, 3, \dots, n$ are the rules, $w_{j}$ denotes the firing strength of jth rule and total number of rules is computed as, $\begin{array}{rcl} R & = & PR (HA 1) * PR (HA 2) \\ * PR (HA 3) * PR (HA 4) \\ * PR (HA 5) \end{array}$ where R is rule count, and $PR (HA)$ is the number of fuzzy sets associated with input variable HA.

Layer 3: This layer normalizes the firing strength of every rule as, $\begin{array}{l} O_{j}^{3} = {\overline{W}}_{j} = \frac{W_{j}}{W_{1} + W_{2} \dots W_{R}}, \\ i = 1, 2, 3, \dots, R \end{array}$

Layer 4: This layer multiplies the normalized firing strength with first order sugeno fuzzy polynomial as follows. $\begin{array}{l} O_{j}^{4} = {\overline{W}}_{j} F_{j} = {\overline{W}}_{j} (p_{1} y + p_{2} y + p_{3}) \\ i = 1, 2, \dots, R \end{array}$ where ${p_{1}, p_{2}, p_{3}}$ denotes the parameter set.

Layer 5: This layer having a single node, which evaluates the summation of all the input signals. $\begin{array}{l} O_{j}^{5} = \sum_{j} {\overline{W}}_{j} F_{j}, j = 1, 2, \dots, R \end{array}$

Layer 1 and 4 use hybrid learning with training data to adjust the parameters. The training process stops when the error tolerance outcome is achieved, or the maximum number of epochs comes. The ANFIS classifies the CHD risk of the user as Normal, Mild, and High. The diagnosis of LDA and ANFIS is combined to determine the health risk of a user, as shown in Algorithm 1. After the determination of the health risk, the alerts are generated immediately to the concerned stakeholders.

Algorithm 1:

LDA-ANFIS based health risk determination

3.2.1.4. Alert generation This module notifies the user about their diagnosed health risk level in real-time. The levels of health risk include i.e. NoR, LoR, MeR, and HiR. If the user is diagnosed with NoR (no health risk), then the user is advised to take preventive measures. If the user is diagnosed with LoR (low health risk), the user is suggested to follow proper medical treatment at hospitals. If the user is diagnosed with MeR (medium health risk), the user is advised to follow medical treatment at the hospitals, and their location and ECG are monitored continuously. If the system observes any abnormalities in heart-related symptoms, instant notifications are sent to the user, and relatives for taking proper care and treatment. If the user is diagnosed with HiR (high health risk), the healthcare service providers are notified for medical emergency, and the user is taken to the hospital and treated under intensive care.

3.2.2. Cloud space

The cloud space employs cloud computing for sensitive information protection, and TNA-based DENF outbreak risk assessment. This space provides scalable global computation resources for area-wide processing of dengue-related data for providing privacy to user-related sensitive information, and identification of DENF affected and risk-prone areas. Each module of the cloud space is explained as follows.

3.2.2.1. Information protection After the real-time analysis and alert generation, the user data (personal and health), diagnosed health risk, generated alerts along with the collected environmental data, are transmitted to the cloud storage, which is present at the cloud space. The user data contains highly sensitive information. The unauthorized disclosure of this information can lead to a situation of panic among people. Henceforth, the information protection module provides privacy to the user data transmitted to the cloud. It functions in two phases viz. InfoFrag and KeyShare. This module transforms the sensitive user data in a form, which prevents unauthorized access. Each phase of the information protection module is explained as follows.

InfoFrag

In this phase, the user data is fragmented (InfoFrag) into two parts, based on the data sensitivity level i.e., LEV-I and LEV-II. LEV-I contains extremely confidential data i.e., personal data viz. name, age, mobile number, sex, and residential address. LEV-II contains the least sensitive level of information viz. dengue-related attributes, and heart-related attributes. The stored data, in the form of a table ‘D’ at the cloud, having a number of attributes $(a_{1}, a_{2}, a_{3}, \dots, a_{n})$ of user ‘ $U_{i}$ ’ is fragmented to appropriate data fragments as, $\begin{array}{l} F_{1} \cup F_{2} = D \end{array}$ where $F_{i}$ is a fragment with following constraints. $\begin{array}{l} F_{i} \cap F_{j} = Ø, i \neq j and F_{i}, F_{j} \in F \end{array}$

KeyShare

After fragmentation, key sharing (KeyShare) mechanism is applied to distribute the highly sensitive attribute values into ‘m’ parts i.e. $v_{1}, v_{2}, v_{3}, \dots, v_{m}$ . These ‘m’ data pieces are stored on secure cloud servers. In this phase, $p - 1$ polynomial coefficients $a_{1}, a_{2}, \dots, a_{p - 1}$ are chosen, and assigned with secret key values to the coefficient $a_{0}$ in order to form a $(p - 1)$ degree polynomial as shows in Eq. (12). $\begin{array}{rcl} K (x) & = & a_{0} + a_{1} x_{1} + a_{2} x_{2} + \dots \\ (12) & + a_{p - 1} x_{p - 1} \end{array}$ where $x_{i}$ is a secret key, chosen from secret keys $X = x_{1}, x_{2}, \dots, x_{n}$ . These key values are chosen randomly, and represents different cloud servers. This module computes each cloud server share by using Eq. (12) and $x_{i} \in X$ . The recreation of the original value of the protected attribute from key share requires knowledge of ‘p’ pieces of data, where $p ⩽ n$ , along with secret key information i.e. $x_{1}, x_{2}, \dots, x_{n}$ . But the secret key is in the possession of the trusted server. Hence, only trusted server has the authority to recreate the original values.

Even if someone accesses the LEV-II information, the revealing of the exact user identity of the infected user cannot be realized. It requires access to both fragments for revealing the user identity of the infected patient. However, the LEV-I data is highly secured.

3.2.2.2. TNA-based DENF outbreak risk assessment The global identification of the dengue infected and risk-prone areas can assess the risk of DENF outbreak, and alert the uninfected users for taking precautions. At the same time, the DENF outbreak risk assessment can alert the government and healthcare agencies for providing preventive measures and required medical support in those areas. The TNA-based module works in this direction of DENF outbreak risk assessment and identification of infected and risk-prone areas. The cloud space employs temporal network analysis (TNA) to assess the DENF outbreak risk. For risk assessment, this module considers the areas, where the residents are infected as infected areas, and the areas, where the density of mosquito is high, as risk-prone areas. TNA analyzes the effect of close proximity for evaluating the susceptance of an uninfected user to catch DENF infection if the user comes in contact with an infected user or mosquito sites. This analysis helps TNA in ascertaining the DENF outbreak rate. Algorithm 2 depicts how the TNA graph is built using the locations of users along with their diagnosed DENF category and the locations of mosquito sites. In the TNA graph, the nodes represent the location of users labeled with their diagnosed DENF categories and the locations of mosquito sites. TNA establishes the edges between the user nodes, which are in close proximity or with mosquito sites. The TNA analyzes the data of GPS sensors of the handheld mobile devices of the users and mosquito sensors to ascertain the close proximities and its durations. The proposed CPS uses Gephi 0.9.1 to create TNA graph.

Algorithm 2:

Temporal network analysis graph

The TNA-based DENF outbreak risk assessment helps in identifying and predicting the infected areas and risk-prone areas. The uninfected users and agencies will be alerted accordingly, and the outbreak will be controlled. The TNA analyzes the temporal network uses the following metrics to assess the DENF outbreak risk.

Temporal Closeness

This metric evaluates the closeness of an uninfected user with the other nodes, as shown in Eq. (13). $\begin{array}{l} {Temporal Closeness}_{user} \\ (13) & = \frac{N - 1}{\sum_{j} DIS (q, r)} \end{array}$ where $DIS (q, r)$ represents the shortest temporal distance of a particular node ‘q’ to every other rth node, and ‘N’ represents the number of nodes in a network. The rth node can be a mosquito site, uninfected user or infected user.

Temporal Path Length (TempPathLen)

This metric determines how fast DENF can be transmitted from mosquito sites and infected users to other users in temporal network. TempPathLen represents the pandemic nature of the DENF outbreak. Lower the value of TempPathLen, faster will be the DENF outbreak. It is evaluated as shown in Eq. (14). $\begin{array}{l} Temporal Path Length \\ (14) & = \frac{\sum_{q r} DIS (q, r)}{N (N - 1)} \end{array}$

Temporal Correlation Coefficient (TempCorCoe)

TempCorCoe is a significant metric to assess the risk of DENF. It identifies the vital DENF clustered regions, and helps in alerting the agencies and users for those regions. This metric evaluates the cluster forming probability of the dengue infected users in a region as show in Eq. (15)–(17). $\begin{array}{l} (15) & C = \frac{1}{N} \sum_{i} C_{q} \\ (16) & C_{q} = \frac{1}{M - 1} \sum_{m = 1}^{m - 1} C_{q} (t_{m}, t_{m - 1}) \\ C_{q} (t_{m}, t_{m - 1}) \\ (17) & = \frac{\sum_{r} a_{q r} (t_{m}) a_{q r} (t_{m + 1})}{\sqrt{[\sum_{r} a_{q r} (t_{m})]} \sum_{r} a_{q r} (t_{m + 1})} \end{array}$ $C_{q} (t_{m}, t_{m - 1})$ represents the sum of neighborhood overlap in the topology around infected node ‘q’ during the time window $(t_{m}, t_{m - 1})$ . $a_{q r}$ represents the direct edge between node ‘q’ and node ‘r’.

Epidemic Relative Score (ERS)

ERS helps the TNA to evaluate the probability of a node in transmitting the DENF infection. This metric is based on the fact that the infected users and mosquito sites in neighborhood contribute in increasing the chances of an infected user in transmitting the DENF infection. Based on this fact, a relative score for every qth node is evaluated as shown in Eq. (18). $\begin{matrix} (18) & {ERS}_{q} = \sum_{r \in V} \sum_{s \in V} \frac{σ_{r s} (q)}{σ_{r s}} \end{matrix}$ where $r \neq s$ , $σ_{r s}$ represents the count for all of temporal shortest paths from ‘r’ to ‘s’. $σ_{j k} (i)$ represents the count for all those temporal shortest paths, which crosses infected node ‘q’.

4. Experimental evaluation

This section discusses the experimental setup and evaluation of the proposed CPS. The entire experimentation and its evaluation is divided into five parts: data collection, LDA-ANFIS classification efficiency, ECG processor efficiency, Alert generation efficiency, and TNA-based DENF outbreak risk efficiency. Each part of the experimental evaluation is explained as follows.

4.1. Data collection

The data is collected from the three government hospitals of the Punjab province in India during the period from 05 September to 27 December of 2018, which was the usual peak period of DENF in India. The dengue-related data consist of ten attributes, namely fever, pain behind eyes, vomiting, severe headache, joint pain, soft bleeding, muscle pain, severe abdominal pain, nausea, and skin rash. For heart-related data, six attributes are collected from the same persons who came to the hospital for DENF investigation. The attributes include namely sex, age, BG, SBP, HDL, and ECG readings. For ECG readings acquisition, the sampling rate was set to 183.60 Hz. During the entire period, 10,321 persons voluntarily provided their health data. After analysis and discussion with the healthcare professionals, 10,000 records are kept based on their variation to provide maximum possible combinations. For capturing ECG signals, the data collection process used 128 amplifier and converted the signals into digital form using 12-bit analog/digital resolution.

4.2. LDA-ANFIS classification efficiency

The efficiency of the classification module LDA-ANFIS is very significant for the proposed CPS to achieve its objectives. The current evaluation uses 3000 user-health records for training the LDA-ANFIS using fuzzy toolbox of Matlab [6]. The training of LDA-ANFIS is iterated using the different number of epochs to minimize the error. The evaluation found that the error was minimized at 50th epoch. After optimized training, the classifier was tested with 7000 user records at 50 epochs. Figure 4(a)–(e) compares the statistical performance of the LDA-ANFIS with other employed classifiers through sensitivity, specificity, accuracy, precision, and F-measure. The evaluation employed Naive Bayes (NB) [36], ANFIS, and Fuzzy System [9]. The results acknowledge the high average sensitivity (93%), accuracy (92%), specificity (89%), F-measure (90%), and precision (90%) of the LDA-ANFIS as compared to the other employed classifiers. The statistical results acknowledge the utility of LDA-ANFIS for health risk classification in the proposed system.

Fig. 4.

Statistical analyses (a) sensitivity of different algorithms (b) specificity of different algorithms (c) accuracy of different algorithms (d) precision of different algorithms (e) F-measure of different algorithms.

Fig. 5.

Original ECG signals and fast Walsh Hadamard coefficients of (a) normal patients (b) abnormal patients.

4.3. Efficiency of ECG signals

This section analyzed the ECG signals in Matlab. The experiment was performed on a 4 GB Windows 10-based Intel i5 machine. Initially, FasWHTrans was applied to decompose the signals. ECG signals consist of various features such as P wave, Q wave, T wave, S wave, R wave, and heart rate. These waves were used to detect a number of heart parameters. Among all, the experiment extracted four intervals RR, PR, QT, QRS from these waves in terms of frequency and spectral domain and pulses to calculate heart rate. Information on heart rate is an instant method for keeping track of heart’s activities. The FasWHTrans decomposed each patient’s ECG data file with 4021 points and generated 8116 coefficients. Figure 5 depicts the original ECG signals and coefficients of FasWHTrans for classifying ECG readings of high-risk patients into a normal and abnormal state. The left side ECG signals in Fig. 5(a) depicts the normal heart condition, and the signal on the right side of the figure represents FasWHTrans coefficients of the normal heart condition. Whereas the left side ECG signals in Fig. 5(b) depicts the abnormal heart condition, and the signal on the right side of the figure represents FasWHTrans coefficients of the abnormal heart condition. After extracting useful features, the Gaussian process model classified the data into two states normal or abnormal ECG readings. The Gaussian process model is implemented on Weka 3.6 [15] with 500 samples to evaluate the performance, and its classification efficiency is represented in Table 3.

Table 3
Classification efficiency of Gaussian process model

Parameter Value

Total Number of Instances 500

Correctly Classified Health Records 92%

Incorrectly Classified Health Records 8%

Kappa Statistic 0.77

Root Relative Squared Error 52.69%

Mean Absolute Error 0.0612

Root Mean Squared Error 0.2354

Relative Absolute Error 11%

Coverage of Cases (0.97 Level) 93%

Parameter	Value
Total Number of Instances	500
Correctly Classified Health Records	92%
Incorrectly Classified Health Records	8%
Kappa Statistic	0.77
Root Relative Squared Error	52.69%
Mean Absolute Error	0.0612
Root Mean Squared Error	0.2354
Relative Absolute Error	11%
Coverage of Cases (0.97 Level)	93%

4.4. Alert generation

The immediate alert generation after determining the health risk of the user is of great significance for the success of the proposed CPS. The efficiency of the alert generation module was evaluated in terms of true and timely alerts generated from CPS. The statistical evaluation of the alert generation is shown in Table 4. The results depict the performance of alert generation with less error rate. For the timely alert generation, the experiment had considered the analysis of delay from the time when the risk was determined to the time when the generated alert was delivered to the user. The analysis found that, on average, each alert took 1.34 seconds to deliver from dew space, which was quite timely.

Table 4
Statistical results of alert generation

Statistical Parameters Values in %

Sensitivity 91

Precision 90.1

Specificity 93.2

Coverage 95.8

Relative Absolute Error 8.18

Root Relative Square Error 32.1

Root Average Square Error 2.81

Mean Absolute Error 2.76

False Positive Alerts 3.48

Statistical Parameters	Values in %
Sensitivity	91
Precision	90.1
Specificity	93.2
Coverage	95.8
Relative Absolute Error	8.18
Root Relative Square Error	32.1
Root Average Square Error	2.81
Mean Absolute Error	2.76
False Positive Alerts	3.48

Table 5

TNA parameters

Average path length	48,121
Network diameter	154
Strongly connected components	180,155
Weakly connected components	31,237
Average weighted degree	2,0000
Average degree	3,0000

4.5. TNA-based DENF outbreak risk assessment

The TNA of the proposed system is created and evaluated using Gephi 0.9.1, as shown in Fig. 6. Table 5 depicts the statistical evaluation of TNA. The evaluation also depicts the analysis of TNA through various metrics viz. eigenvector centrality and betweenness centrality distribution, as shown in Fig. 7. The eigenvector centrality depicts the ERS of various user nodes. The betweenness centrality distribution identifies the nodes which act as a bridge between nodes and indirectly contributes to the DENF infection outbreak. The analysis of TNA metrics helps in assessing the DENF outbreak risk through the identification of critical areas, and users, which are probable for spreading DENF infection.

Fig. 6.

DENF-based TNA graph.

Fig. 7.

TNA metrics (a) eigenvector centrality distribution, and (b) betweenness centrality distribution.

5. Conclusion

In this paper, dew-cloud assisted CPS is proposed for real-time diagnosis of DENF, and monitoring of CHD risk of DENF infected users. The proposed system addresses the issue concerning the effects of DENF infection on the vital body organs at the dew space, through real-time monitoring of CHD risk during the DENF infection period. The system uses dew computing for real-time determination of the health risk of the users based on their DENF infection and CHD risk level, immediate diagnostic and warning alert generation to the concerned stakeholders and timely provision of medical support during any dengue-related emergency. At cloud space, the proposed system provides information protection against unauthorized disclosure of users’ sensitive information and TNA-based DENF outbreak risk assessment for identifying the critical areas and users, who contribute in spreading the DENF infection. All the aspects regarding the diagnosis, monitoring, and depiction of DENF risk are organized in the proposed intelligent healthcare system. The experimental evaluation of the system is based on the health records of 10,000 patients. The efficiency evaluation of the various modules justifies the utilization of the proposed system. The statistical results comparison of LDA-ANFIS based health risk classification with other employed classifiers, the preprocessing of ECG and its performance analysis, alert generation performance, and TNA-based DENF outbreak risk assessment through Gephi acknowledges the performance efficiency and utilization of the proposed system. The proposed system in the current paper just considered the effect of DENF on one vital organ i.e., heart, but the system is open for future extensions to other vital organs’ consideration too, for DENF infection and other alike diseases.

References

Ahmad ,

M.B.

Amin ,

Hussain ,

B.H.

Kang ,

Cheong and

Lee , Health fog: A novel framework for health and wellness applications, The Journal of Supercomputing 72(10) (2016), 3677–3695. doi:10.1007/s11227-016-1634-x.

Andriopoulou ,

Dagiuklas and

Orphanoudakis , Integrating IoT and fog computing for healthcare service delivery, in: Components and Services for IoT Platforms, Springer, 2017, pp. 213–232. doi:10.1007/978-3-319-42304-3_11.

Arora and

R.S.

Patil , Cardiac manifestation in dengue fever, J Assoc Physicians India 64(7) (2016), 40–44.

Axak ,

Rosinskiy ,

Barkovska and

Novoseltsev , Cloud-fog-dew architecture for personalized service-oriented systems, in: 2018 IEEE 9th International Conference on Dependable Systems, Services and Technologies (DESSERT), IEEE, 2018, pp. 78–82. doi:10.1109/DESSERT.2018.8409103.

Bogdan ,

Jain ,

Goyal and

Marculescu , Implantable pacemakers control and optimization via fractional calculus approaches: A cyber-physical systems perspective, in: 2012 IEEE/ACM Third International Conference on Cyber-Physical Systems, IEEE, 2012, pp. 23–32. doi:10.1109/ICCPS.2012.11.

Bystrov and

Westin , Practice. Neurofuzzy Logic Systems, Matlab Toolbox Gui, 8–39.

Cardiovascular diseases (CVDs).

Chakraborty ,

S.K.

Ghosh ,

Jamthe and

D.P.

Agrawal , Detecting mobility for monitoring patients with Parkinson’s disease at home using RSSI in a wireless sensor network, Procedia Computer Science 19 (2013), 956–961. doi:10.1016/j.procs.2013.06.132.

C.-H.

Cheng and

D.-L.

Mon , Fuzzy system reliability analysis by interval of confidence, Fuzzy Sets and Systems 56(1) (1993), 29–35. doi:10.1016/0165-0114(93)90182-H.

10.

Costanzo ,

Faro ,

Giordano and

Pino , Mobile cyber physical systems for health care: Functions, ambient ontology and e-diagnostics, in: 2016 13th IEEE Annual Consumer Communications & Networking Conference (CCNC), IEEE, 2016, pp. 972–975.

11.

Dey ,

A.S.

Ashour ,

Shi ,

S.J.

Fong and

J.M.R.

Tavares , Medical cyber-physical systems: A survey, Journal of medical systems 42(4) (2018), 74. doi:10.1007/s10916-018-0921-x.

12.

Y.A.

Geadah and

Corinthios , Natural, dyadic, and sequency order algorithms and processors for the Walsh–Hadamard transform, IEEE Transactions on Computers C-26(5) (1977), 435–442. doi:10.1109/TC.1977.1674860.

13.

J.T.

Giger ,

N.D.

Pope ,

H.B.

Vogt ,

Gutierrez ,

L.A.

Newland ,

Lemke and

M.J.

Lawler , Remote patient monitoring acceptance trends among older adults residing in a frontier state, Computers in Human Behavior 44 (2015), 174–182. doi:10.1016/j.chb.2014.11.044.

14.

Gusev , A dew computing solution for IoT streaming devices, in: 2017 40th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO), IEEE, 2017, pp. 387–392. doi:10.23919/MIPRO.2017.7973454.

15.

Hall ,

Frank ,

Holmes ,

Pfahringer ,

Reutemann and

I.H.

Witten , The WEKA data mining software: An update, ACM SIGKDD explorations newsletter 11(1) (2009), 10–18. doi:10.1145/1656274.1656278.

16.

S.A.

Haque ,

S.M.

Aziz and

Rahman , Review of cyber-physical system in healthcare, International Journal of Distributed Sensor Networks 10(4) (2014), 217415. doi:10.1155/2014/217415.

17.

He ,

Fan and

Li , Toward ubiquitous healthcare services with a novel efficient cloud platform, IEEE Transactions on Biomedical Engineering 60(1) (2012), 230–234. doi:10.1109/TBME.2012.2222404.

18.

S.-Y.

Huang ,

Lee ,

Wang ,

J.-W.

Liu ,

S.-C.

Hung ,

C.-C.

Chen ,

T.-Y.

Chang and

W.-C.

Huang , Use of simple clinical and laboratory predictors to differentiate influenza from dengue and other febrile illnesses in the emergency room, BMC infectious diseases 14(1) (2014), 623. doi:10.1186/s12879-014-0623-z.

19.

Ibrahim ,

Thio ,

Faisal and

Neuman , The application of biomedical engineering techniques to the diagnosis and management of tropical diseases: A review, Sensors 15(3) (2015), 6947–6995. doi:10.3390/s150306947.

20.

M.A.

Karaolis ,

J.A.

Moutiris ,

Hadjipanayi and

C.S.

Pattichis , Assessment of the risk factors of coronary heart events based on data mining with decision trees, IEEE Transactions on information technology in biomedicine 14(3) (2010), 559–566. doi:10.1109/TITB.2009.2038906.

21.

Kularatne ,

Rajapakse ,

Ralapanawa ,

Waduge ,

Pathirage and

Rajapakse , Heart and liver are infected in fatal cases of dengue: Three PCR based case studies, BMC infectious diseases 18(1) (2018), 681. doi:10.1186/s12879-018-3603-x.

22.

Lyu ,

Hong ,

Wei ,

Yang ,

Tang ,

Wang and

Agoulmine , Dynamic evaluation model of coronary heart disease for ubiquitous healthcare, Computers in Industry 69 (2015), 35–44. doi:10.1016/j.compind.2014.09.008.

23.

Mendez and

M.C.

Van den Hof , Mobile remote-presence devices for point-of-care health care delivery, CMAJ 185(17) (2013), 1512–1516. doi:10.1503/cmaj.120223.

24.

Paoletti ,

Jiang ,

M.A.

Islam ,

Abbas ,

Mangharam ,

Lin ,

Gruber and

S.A.

Smolka , Synthesizing stealthy reprogramming attacks on cardiac devices, in: Proceedings of the 10th ACM/IEEE International Conference on Cyber-Physical Systems, ACM, 2019, pp. 13–22.

25.

T.F.

Póvoa ,

A.M.

Alves ,

C.A.

Oliveira ,

G.J.

Nuovo ,

V.L.

Chagas and

M.V.

Paes , The pathology of severe dengue in multiple organs of human fatal cases: Histopathology, ultrastructure and virus replication, PloS one 9(4) (2014), e83386. doi:10.1371/journal.pone.0083386.

26.

Prittopaul ,

Sathya and

Jayasree , Cyber physical system approach for heart attack detection and control using wireless monitoring and actuation system, in: 2015 IEEE 9th International Conference on Intelligent Systems and Control (ISCO), IEEE, 2015, pp. 1–6.

27.

I.A.

Rather ,

H.A.

Parray ,

J.B.

Lone ,

W.K.

Paek ,

Lim ,

V.K.

Bajpai and

Y.-H.

Park , Prevention and control strategies to counter dengue virus infection, Frontiers in cellular and infection microbiology 7 (2017), 336. doi:10.3389/fcimb.2017.00336.

28.

P.P.

Ray , An introduction to dew computing: Definition, concept and implications, IEEE Access 6 (2017), 723–737. doi:10.1109/ACCESS.2017.2775042.

29.

P.P.

Ray ,

Dash and

De , Internet of things-based real-time model study on e-healthcare: Device, message service and dew computing, Computer Networks 149 (2019), 226–239. doi:10.1016/j.comnet.2018.12.006.

30.

M.J.

Reddy and

Kavitha , Expert System to Predict the Type of Fever Using Data Mining Techniques on Medical Databases, 2015.

31.

Rindos and

Wang , Dew computing: The complementary piece of cloud computing, in: 2016 IEEE International Conferences on Big Data and Cloud Computing (BDCloud), Social Computing and Networking (SocialCom), Sustainable Computing and Communications (SustainCom) (BDCloud-SocialCom-SustainCom), IEEE, 2016, pp. 15–20. doi:10.1109/BDCloud-SocialCom-SustainCom.2016.14.

32.

Samanta and

Sharma , Dengue and its effects on liver, World Journal of Clinical Cases: WJCC 3(2) (2015), 125–131. doi:10.12998/wjcc.v3.i2.125.

33.

A.T.

Thakar and

Pandya , Survey of IoT enables healthcare devices, in: 2017 International Conference on Computing Methodologies and Communication (ICCMC), IEEE, 2017, pp. 1087–1090. doi:10.1109/ICCMC.2017.8282640.

34.

T.-L.

Thein ,

E.-L.

Ng ,

M.S.

Yeang ,

Y.-S.

Leo and

D.C.

Lye , Risk factors for concurrent bacteremia in adult patients with dengue, Journal of microbiology, immunology and infection 50(3) (2017), 314–320. doi:10.1016/j.jmii.2015.06.008.

35.

Tunmibi ,

Adeniji ,

Aregbesola and

Dasylva , A rule based expert system for diagnosis of fever, International Journal of Advanced Research 1(7) (2013), 343–348.

36.

G.Y.

Tütüncü and

Kayaalp , An aggregated fuzzy naive Bayes data classifier, Journal of Computational and Applied Mathematics 286 (2015), 17–27. doi:10.1016/j.cam.2015.02.004.

37.

Vikram ,

Nagpal ,

Pande ,

Srivastava ,

Saxena ,

Anvikar ,

Das ,

Singh ,

S.K.

Gupta ,

Tuli et al., An epidemiological study of dengue in Delhi, India, Acta tropica 153 (2016), 21–27. doi:10.1016/j.actatropica.2015.09.025.

38.

H.U.H.

Virk ,

Inayat and

Z.U.

Rahman , Complete heart block in association with dengue hemorrhagic fever, Korean circulation journal 46(6) (2016), 866–869. doi:10.4070/kcj.2016.46.6.866.

39.

World Health Organization, Dengue and severe dengue, 2019, Accessed: 2020-01-14.

40.

Xia ,

Asif and

Zhao , Cloud-ECG for real time ECG monitoring and analysis, Computer methods and programs in biomedicine 110(3) (2013), 253–259. doi:10.1016/j.cmpb.2012.11.008.

41.

Xu ,

Da Xu ,

Cai ,

Xie ,

Hu and

Bu , Ubiquitous data accessing method in IoT-based information system for emergency medical services, IEEE Transactions on Industrial informatics 10(2) (2014), 1578–1586. doi:10.1109/TII.2014.2306382.

42.

J.-G.

Yang ,

J.-K.

Kim ,

U.-G.

Kang and

Y.-H.

Lee , Coronary heart disease optimization system on adaptive-network-based fuzzy inference system and linear discriminant analysis (ANFIS-LDA), Personal and Ubiquitous Computing 18(6) (2014), 1351–1362. doi:10.1007/s00779-013-0737-0.

43.

Zhang ,

Liu ,

Tsang and

Virginas , Expensive multiobjective optimization by MOEA/D with Gaussian process model, IEEE Transactions on Evolutionary Computation 14(3) (2009), 456–474. doi:10.1109/TEVC.2009.2033671.

44.