CHAPTER 2

LITERATURE SURVEY

2.1 OUTLINE

This chapter aims to provide the recent works that are carried out on different prediction
models which are aimed at improving the accuracy and efficacy of the prediction. It provides
a detailed survey along with the classification from works related to early prediction, at the
affected stage and after the stage. Also, it has been segregated based on supervised classifiers,
unsupervised classifiers and deep learning prediction models.
The presence of a symptom of a heart disease and mortality rate has been increasing
day-by-day throughout the different communities of the world. CVD is one of the most
important diseases that play a vital role in medical decision support systems. Healthcare
industries collect huge sums of data in order to ascertain or take business decisions and
predictions. The advent of data mining and machine learning techniques proved its
importance which can help the physicians to make informed decisions and predictions.
Several factors contribute to heart disease such as diabetes, high blood pressure, high
cholesterol, abnormal pulse rate, age, sex, location, etc.
Machine learning proved to be effective in diagnosing or predicting the presence or
absence of heart disease in advance. Researchers have utilized various different datasets
ranging from real-time to storage bases like, UCI repository, Cleveland Dataset, Hungarian
Datasets and so on. Machine learning algorithms can be broadly classified into supervised and
unsupervised learning methods. In order to improve the prediction and diagnosing of heart
disease several research works have been contributed by various researchers. Depending upon
the attributes or features different parameters such as accuracy, specificity and sensitivity are
utilized to calculate the efficacy of the techniques implemented. According to the author,
the study has been conducted using various computing platforms such as Elastic Compute
Cloud (EC2) Amazon Cloud, Apache Spark, Waikato Environment for Knowledge Analysis
(WEKA) tools, etc. To assess and to improve the efficacy of the system, validation is being
carried out by using Sequential Forward Selection (SFS), K-fold cross validation, and Relief
algorithms, Region Operational Curves (ROC).
Section 2.2 represents the research work carried out to predict the heart disease via
hybrid machine learning classification methods pertaining to different environments. Section

18
2.3 depicts various works related to prediction of heart disease via unsupervised learning
methods. Section 2.4 depicts various works published utilizing machine learning algorithms
to detect heart disease using wearable devices and section 2.5 represents a table of research
works published.

2.2 HEART DISEASE PREDICTION USING HYBRID MACHINE LEARNING

CLASSIFICATION ALGORITHMS

Most of the research works that have been carried out utilized both supervised and
unsupervised machine learning algorithms in order to diagnose and to predict CVD. In
Supervised machine learning, the model is trained and tested using the class labels where the
input and the output are the class labels. In the real-world, supervised learning can be used
for Risk Assessment, Image classification, Fraud Detection, spam filtering, etc. Supervised
machine learning can be categorized into two types namely Prediction and Classification.
Regression algorithms are used if there is a relationship between the input variable and
the output variable. It is used for the prediction of continuous variables, such as Weather
forecasting, Market Trends, etc. There are different kinds of regression-based methods
such as linear regression, regression trees, Non-linear regression, polynomial regression, etc.
Classification algorithms are used when the output variable is categorical, which means there
are two classes such as Yes-No, Male-Female, True-false, etc. Some of the classification
based supervised machine learning algorithms are random forests, decision trees, logistic
regression, SVM, etc. Few authors have utilized clustering and Principal Component Analysis
(PCA) techniques in order to analyze the performance of the algorithms used. A detailed
survey of different works carried out by varied authors across different zones on different
datasets. The study has been extensively carried out in order to improve the accuracy rather
than specificity and sensitivity.
Atherosclerotic Coronary Heart Disease (ACHD) has been a major cause for the higher
mortality rate since the 20th Century. Kumar and Gandhi [2018] conducted a machine
learning based investigation in order to find the outcomes of the least influential medicinal
practice involved in detecting heart disease and to find the impact of deficiencies relating to
data of the patients. Amazingly, the use of supervised machine learning techniques proved
to be efficient in detecting the new insights with respect to the non-invasive medical practice
and the effect of machine learning technique in detecting the atherosclerotic coronary heart

19
disease.
Magnetocardiography has been utilized to detect the irregularity or the restriction of
blood flow to the tissues of the heart. Tantimongcolwat et al., [2008] proposed that detecting
the ischemic condition at an early stage provides a high possibility of reducing or alleviating
the mortality rate. Though magneto cardiograms are capable of detecting the abnormal
patterns of an ICD, analyzing to find the abnormality requires high-time and a highly
trained professional. The authors have chosen two machine learning approaches such as
Back-Propagation Neural Network (BPNN) and Direct-Kernel Self Organizing Mapping
technique (DK-SOM). Data sets have been trained and tested by using both these approaches,
where DK-SOM proves more efficient in obtaining higher sensitivity, specificity and accuracy
than BNN. Machine learning approaches prove to be more efficient and the detection can be
done at a faster pace than the other methods.
Kangwanariyakul et al., [2010], performed an experiment by comparing the machine
learning models in order to detect the Ischemic condition pertaining to heart disease. Globally,
IHD causes a high mortality rate. Magnetocardiography being an efficient tool which is a
non-invasive method requires a trained professional and it takes time to diagnose. Hence to
identify whether a patient has abnormal flow of blood to the heart, the author performed a
comparative study by using techniques such as Back-Propagation Neural Network (BPNN),
Bayesian Neural Network (BNN), Probabilistic Neural Network (PNN) and the Support
Vector Machine (SVM). After the analysis, BPNN and BNN proved to be more efficient in
the earlier detection of IHD among patients. Linear SVM proves to be the least which shows
a specificity of 45 % and polynomial SVM shows highest specificity of 86 %.
Scientists have utilized several data mining techniques in order to help the physicians in the
early diagnosis of heart disease. Shouman et al., [2011] proposed that the decision tree being
an efficient method in combination with voting provides a high pace and accuracy. Though
the supervised discretization methods do not prove to be efficient, engagement of the voting
technique provides high efficiency in the detection of heart disease while employing decision
trees and gain ratio. The method employed outperformed the traditional J4.8 Decision trees
and the Bagging Algorithm.
For the earlier diagnosis of heart disease; K-Nearest Neighbor algorithm (KNN) being a
classifier model being least used for the prediction of heart disease in an individual. Shouman
et al., [2012], have identified that the classifier model in combination with the voting method
improved the level of accuracy. In order to investigate the performance of KNN integrated

20
with the voting method a detailed research study has been conducted. The results prove that
KNN integrated with the neural network ensemble provides more accuracy than KNN with
a voting method.
Due to the advancements in technology and its applications, scientists quite often face
difficulty in making medical decisions. Though artificial neural networks provide more
promise in the prediction of heart disease, due to the complex nature of the neural learning
systems its generality has been questioned. Hence in order to provide a solution in case
of patients pertaining to different illnesses, SVM integrated with Sequential Minimize
Optimization (SMO) technique has been proposed. Parameters such as accuracy, sensitivity
and specificity prove to be more efficient than the other conventional approaches.
Parthiban [Link] Parthiban and Srivatsa [2012] comes up with a query that the presence
of diabetes indicates the absence of production of insulin from the pancreas. Heart disease
can also be found in diabetic patients too. Hence the detection of heart disease in diabetic
patients facilitates earlier diagnosis to avoid it. Techniques such as NB and SVM provide
a most promising solution toward the prediction of heart disease. Utilizing SVM to detect
heart disease in diabetic patients shows better results with high accuracy.
Data mining has potential in providing solutions across different applications. In medical
care applications, data mining techniques help in the prediction of diseases thereby facilitating
the way for the cure. Vijayarani and Sudha [2013] proposed comparative work by analyzing
the efficiency of logistic regression, Multi-Layered Perceptron (MLP) and SMO algorithms.
The analysis has been performed using the WEKA tool. Clustering accuracy and error rate
are chosen as the parameters for analyzing its efficiency of the algorithms. The study proves
that the logistics function provides better efficiency than the MLP and SMO.
Tomar and Agarwal [2014] came up with a Feature Selection based Least Twin Support
Vector Machine (LSTSVM) is an approach used in the machine learning technique. F-Score
being used as a feature selection metric used to weigh each feature. Grid approach is used to
evaluate the performance of the parameters. Results provide a most evident proof that outlines
all the traditional approaches proposed earlier. The proposed model works efficiently only on
the classification of two classes. In case of more classes, feature selection and optimization
techniques can also be utilized further.
CHD has been more prevalent among the aged and the young adults nowadays. Any
treatment involves two major tasks: diagnosing and monitoring. Ahmed et al., [2015]
proposed a solution based on a machine learning technique called SVM. For monitoring

21
the system includes a wearable component made up of IoT. These devices facilitate efficient
monitoring of patients even from a far remote place. IoT wearable devices when connected
with the internet enable early detection and diagnosis for the elderly in case of emergency
services. For diagnosing the system employs a test and a training set from 20 patients and
three rounds of tests have been taken in order to analyze the performance of SVM. To evaluate
this, SVM technique is compared with BayesNet and DT. 10-fold cross validation technique
has been employed to prove the effectiveness of the classifier employed. In the third round,
feature selection has been employed in order to achieve higher accuracy. The result proves
that the employed SVM technique provides an accuracy of about 86 %.
Patel et al., [2016a], in order to reduce the deaths due to heart disease, the author proposes
that the decision tree may provide a most promising solution. Different decision tree based
algorithms such as J48, Logistic model tree algorithm and the RF algorithm. By using
the WEKA tool, the research has been conducted in order to evaluate the performance of
different algorithms. In the study J48 provides evident results which have achieved only
marginal success. Still it suffers from errors due to pruning. Hence combination of different
models and algorithms may prove its accuracy.
Prediction of heart disease by using optimization techniques oppositional firefly with Bat
algorithm and Rule-Based Fuzzy Logic (RBFL). Automation of these models will promote
efficient diagnosis of disease. Reddy and Khare [2017] proposed a hybrid Oppositional
Firefly-Based Algorithm - Rule-Based Fuzzy Logic (OFBAT-RBFL) diagnosing system
where the features are selected using the Locality Preserving Projection (LPP) algorithm.
To help in diagnosing a fuzzy logic OFBAT algorithm to improve the BAT algorithm has
been utilized. The fuzzy system encompasses certain rules and membership functions which
outperforms the existing traditional approaches to an accuracy of 78%.
There are several factors that contribute to the development of coronary heart disease.
Almost one-fourth of every person dies due to coronary heart disease. Miao et al., [2016]
proposed a research study based on adaptive boosting algorithms in order to predict the
presence and the absence of coronary heart disease in an individual. The results have been
trained and tested using the four different hospital data obtained from the Hungarian patients
where the developed model provides higher accuracy than the other conventional approaches.
The ensemble classifier and the prediction model proposed are evaluated with two different
classes such as the presence and the absence of CHD.
Cleveland dataset has been used. Gupta et al., [2017] have conducted a research study

22
through ML and cloud computing platforms. Since cloud computing promotes the delivery
of on-demand resources, it plays a vital role in health care too. In comparison with AdaBoost
and boosted tree methods Naive bayes have produced a higher accuracy rate of 88%. The
results have been evaluated with nearly 10,082 instances within a minimum execution time
which can be utilized in rural emergency care services.
Tao et al., [2018] conducted a research study in order to predict the IHD in patients.
Several works have already been carried out by various researchers; the authors have carefully
selected various features such as time-domain, frequency-domain and information theory etc.
A comparison among different approaches starting from KNN, DT, SVM and Boost. Authors
found that SVM-XGBoost proved more accurate in predicting and diagnosing the Ischemic
condition in heart disease of about 94 %. Localizing the presence of ischemic conditions in
the heart is the challenging part where it can be easily done by using the magneto cardiograms
efficiently.
A hybrid Intelligent Machine learning based predictive system has been proposed in
order to diagnose CHD which has been on the rise since the past decade Haq et al., [2018]
conducted a detailed research study with comparison on different classifiers such as KNN,
Artificial Neural Network (ANN), SVM, NB and Decision Trees. Three feature selection
algorithms such as relief, minimum Redundancy Maximum Relevance (mRMR) and Least
Absolute Shrinkage and Selection Operator (LASSO) have been utilized in order to determine
the appropriate features used for training and testing the classification model. In order to
validate the research study conducted, a k-fold cross validation test has been performed.
In order to achieve high accuracy logistic regression outliers all the other approaches; in
terms of achieving the specificity SVM with mRMR feature selection-based yields higher
performance. In case of sensitivity, ANN provides 100 % accuracy. Since additional feature
selection algorithms have been used it becomes overhead whereas inclusion of more features
can be a future study which has been indicated by the authors.
Heart diseases are becoming more prevalent nowadays than the olden days. This may
be due to several factors such as age, sex, food habits, climate and other economic factors.
Several ML techniques have been proposed in order to predict and to treat heart disease
with the advent of new technology. Among these techniques, Ashok et al., [2017] has 8
diversified classification attributes in order to find the most accurate technique that would
produce sensitivity, specificity and accuracy. From the analysis, it has been ascertained that
logistic regression performs much better other than ANN, SVM, Naive Bayes, etc.

23
 CHD can be predicted by the deep learning algorithms more efficiently. Diller [Link] Diller
et al., [2019] utilized convolutional neural networks as a training algorithm in order to predict
heart disease. A record of 44,000 instances has been taken into account where the data are
categorized into diagnosis, disease complexity which provides an accuracy of 91 %.
Most of the work that has been proposed in the prediction of coronary heart disease
involves comparison of different machine learning methods. Du et al., [2020] concluded
a research study for the prediction of CHD in hypertension patients. The study includes
non-linear models such as RF, KNN, DT, SVM and XGBoost algorithms. The work has
been performed by using big data and the machine learning methods where the logistic
regression models have been outperformed by these non-linear models which improves the
risk prediction with the high amount of dataset.
Patel et al., [2016b] investigated whether hypertension patients have been affected by
CVD. The investigation involves the ensemble learning classifiers using the majority voting
rule. The approach suffers from a major disadvantage that it has not been tested with large
datasets which makes its validation being a question mark.
Hager et al., [2020] proposed an optimal machine learning algorithm which is univariate
and relief has been chosen. Four types of machine learning algorithms such as DT, SVM, RF
and logistic regression classifiers are utilized to find the accuracy in predicting heart disease.
Validation of these comparisons is done through hyper parameter tuning and cross validation
techniques using the Apache spark tool. It is evident that the RF classifier outlines the other
traditional models with a high accuracy of about 95 %.
Heart disease diagnosis using invasive and ordinary methods are so tedious. Hence in
order to check the effectiveness of different machine learning classifiers six were chosen
and compared with the deep learning algorithm called Bayesian Probability Neural Network
(BPNN). The feature selection algorithm chosen was relief where ensemble classifiers such
as boosting, bagging and stacking have also been utilized. BPNN does not need a feature
selection algorithm. Through the use of ensemble learning classifiers, SVM and BPNN
proves high accuracy. It is inferred from the study that the deep learning algorithm provides
more accuracy than the conventional machine learning approaches.
Clinical decision support systems in terms of predicting heart disease has been a challenge
since the last decade. Nilashi et al., [2020] investigated and proposed a machine learning
technique by using supervised and unsupervised classification techniques. The research
involves Principal Component Analysis (PCA), Self-Organizing Map (SOM), Fuzzy Support

24
Vector Machine (FSVM) are used to reduce the time consumption required during the
diagnosis of heart disease.
Different Machine learning algorithms have been utilized for the prediction of heart
disease. ANN is a classic artificial intelligence algorithm which is combined with many ML
and DL techniques. In order to facilitate the research study conventional machine learning
methods such as random forest, support vector machine, K-nearest neighbors are compared
along with the deep-learning methods such as Long Short-Term Memory (LSTM) and
Gradient Recurrent Unit (GRU) neural networks. In order to validate a majority voting-based
model is used to test the accuracy of the classifiers utilized. It has been identified that there
is a significant increase of 2.1 % in accuracy than the conventional approaches.
In order to ascertain the accuracy of the predictive model, effort estimation is an important
tool used in the software development practice. Singh et al., [2020] combined three different
approaches such as Linear Regression (LR), MLP and RF are utilized and implemented using
the WEKA tool kit, where it is evident that the linear regression outperforms all the other
approaches used.
Aggrawal and Pal [2021] addressed the prediction of death in case of heart disease. The
feature discrimination strategy helps us to reduce computation time, to increase prediction
diagnose time and to have an increased understanding and its applications. In order to detect
the death events of heart disease patient’s different machine learning algorithms such as LDA,
RF, GB Classifier, DT, SVM and KNN are used. The precision of the comparative study
has been analyzed by using the SFS algorithm wherein the RF classifier achieves a higher
accuracy of about 87%.
CHD has been more prevalent which affects different age groups all around the world.
In the last decade, it has been evident that heart attack occurs without any transparent
symptoms; hence an advanced intelligent detection methodology is of prime importance.
A two-tier ensemble classifier is built with the help of three classifiers such as RF, GB,
and XGBoost method. The methodology has been implemented by using different datasets
such as Z-Alizadeh sani, Statlog, Cleveland and Hungarian where the evaluation has been
validated by using the 10-fold cross validation. The system produced higher results when
compared with the previous other conventional approaches.
In clinical decision-making machine learning classifiers with high accuracy are of prime
importance. Ali et al., [2021] utilized different classifiers which are compared where feature
selection was not performed except for MLP and KNN. Hence it is evident that KNN, DT and

25
RF achieved 100 % accuracy, specificity and sensitivity in order to detect the heart disease.
Jinny et al., [2021], identification of a threat relating to heart care is of prime importance
in sufferers with other interlinked diseases such as glucose secretion, increased pressure
changes in the blood, huge concentration of fats, etc. This opens an avenue to devise a
forecasting method for increased threat of patients thereby making their life saved.
Many chronic respiratory diseases such as asthma, COPD, pneumonia, respiratory
distress, etc. are considered to be a significant public health burden. To reduce the mortality
rate, it is better to perform early prediction of respiratory disorders and treat them accordingly.
To build an efficient prediction model for various types of respiratory diseases, machine
learning approaches are used. Priya et al., [2020b] proposed a methodology to build a
classifier model using supervised learning algorithms such as RF, DT, and Multi-Layer
Perceptron Neural Network (MLP-NN) for the detection of different respiratory diseases of
ICU admitted patients. It achieves accuracy of nearly 99 % by various machine learning
approaches.
Priya et al., [2020a] proposes a novel multi-tiered and hybrid architectural framework
to evaluate various learning algorithms for building an early heart disease prediction model.
The most important aspects of machine learning are the ability to capture non-linear
features and it led to a greater assurance on the dataset in order to obtain more accurate
prediction results. The aim of the work is to design the prediction system with few
numbers of features with increased efficiency. To reduce the diagnostic barriers in the health
prediction model, an efficient method of Bayesian optimization for hyper parameter tuning and
k-fold cross-validation resampling technique is applied to evaluate various conventional and
ensemble machine learning techniques. The experimental results determine the efficiency and
validity of the proposed model for heart disease prediction. Also, comparison analysis of the
proposed model with other existing expert systems is carried out to depict better performance.
Finally, our prediction results show a high accuracy of 86.47 % using the Logistic Regression
algorithm and it clearly indicates every classifier model’s accuracy is improved to a greater
extent after applying the Bayesian Optimization and 10-Fold Cross-Validation for heart
disease.

26
2.3 HEART DISEASE PREDICTION USING UNSUPERVISED MACHINE
LEARNING ALGORITHMS

Unsupervised machine learning is one of the types of machine learning approach where facts
are drawn from different datasets that constitute input data with no class labels. Unsupervised
machine learning techniques are broadly classified into two types such as Clustering and
Association. There are different types of unsupervised machine learning algorithms such as
K-Means Clustering, Singular Value Decomposition (SVD), PCA, KNN, Apriori Analysis,
Neural Networks, Anomaly Detection, etc. K-means clustering and association rule learning
algorithms are the commonly used algorithms in unsupervised learning approaches.
Pandey et al., [2013] proposed an attribute selection method in order to predict the heart
disease using the clustering technique. Many different types of clustering techniques such
as density-based cluster, model-based cluster, grid-based cluster and partition-based cluster,
etc. Author utilized a minimum number of instances for training and testing the proposed
method. A comparative study has been performed between K-means and density based
clustering techniques where the density based clustering technique stood high in accuracy of
about 82%.
Kumar et al., [2014], defined a methodology based on K-Means Clustering for disease
detection. The study enhanced the existing Euclidean formula where the clusters are
calibrated. The work has been preceded with the Human Immunodeficiency Viruses (HIV)
dataset and it produced an efficiency of 96 %. However, future directions can be carried out
by extending it for the heart disease prediction.
Heart attack potentially being a deadly disease has caught a deeper attention across the
wide category of researchers all around the world. Ziasabounchi and Askerzade [2014]
proposed a PCA based clustering method based on C-means and K-means. The study was
performed by using PCA and without PCA. K-means via PCA clustering provides a deeper
understanding and prediction of heart disease with a high accuracy of about 97 %
Stricker et al., [2013], conducted a study on dietary patterns which has the associations
with CHD and Stroke. The dietary pattern of different patients has been analyzed by using
PCA and K-means Clustering techniques. Results show that there is eminent evidence that
proves there is a strong association with CHD and stroke. Therefore, a genuine calibrated
dietary pattern would reduce the risk of CHD and stroke has been suggested.
Mattioli et al., [2020] proposed a solution to reduce and to understand the symptoms

27
pertaining to heart disease. Authors aimed at the utilization of dimensionality reduction
method and feature selection for treating the heart patients. Chi-Square and PCA with RF
have been utilized to perform the comparison and validation of the six machine learning
classifiers used. The work has been carried out by using three varying datasets such as
Cleveland, Hungarian and Cleveland-Hungarian datasets. PCA based method Chi-square
method showed higher efficiency than the other types of classifiers.
Islam et al., [2020] presented a research work for the early prediction of heart disease
where in principal component analysis technique has been used to reduce the attributes and
a hybrid genetic algorithm based on k-means clustering has been utilized. The proposed
methodology has produced eminent results with an accuracy of 94 %
Shrestha et al., [2021] proposed a cluster based decision tree learning algorithm where
the Cleveland dataset has been used. Since most of the datasets are based on class labels, use
of machine learning in case of large datasets with class labels becomes inefficient. Hence
in order to achieve higher accuracy Cluster-based Decision Tree Learning (CDTL) based
approach along with Random Forest yields better accuracy of about 89%.

2.4 HEART DISEASE PREDICTION USING WEARABLE DEVICES

PERTAINING TO RURAL ENVIRONMENT

Most of the reviewed papers have been proposed machine learning prediction models
combined with or without other related intelligence techniques for chronic disease detection
of elderly patients at an early stage to avoid emergency situations.
IoT being cost-effective, efficient where any devices can be connected through the internet.
IoT enables the real-time tracking of a patient’s health continuously. Therefore, the wearable
devices utilized in IoT health monitoring systems generate a high and a continuous amount
of data thereby facilitating the health care officials to monitor and to treat the old age people
or any individual in a real-time manner.
Technology pertaining to wearable devices promotes care and course of action pertaining
to patients in medical institutions. Scientists utilize wearable devices for the forecasting
of dissimilarities in health. It enables the development of a new breed of POC diagnostic
devices. For example, garments integrated with wearable solutions, such as commercial
portable sensors and devices in the Emergency Medical Services (EMS), Emergency Room
(ER) or Intensive Care Unit (ICU) environments, have facilitated the continuous monitoring

28
of risks that endanger patient lives. The system enables detection of patient health-state
parameters (heart rate, breathing rate, body temperature, blood oxygen saturation, position,
activity and posture) and environmental variables (external temperature, presence of toxic
gases, and heat ux passing through the garments) to process data and remotely transmits
useful information to healthcare providers.
Wireless wearable devices have supported mobility in patients. Activity monitoring is
used to manage chronic conditions of patients. Wearable device activity tracking abilities
provide a mechanism to allow health consumers to enhance their self-management capacities.
Many health consumers are already tracking their weight, diet, or health routines in some
way. Wearable devices further improve the self-tracking ability by providing sensor data as
objective evidence.
Wearable technology can assess patient heart activity outside of a laboratory or clinical
environment. It is possible to perform heart assessments during a wide range of everyday
conditions without interfering with a patient’s activity tasks. For example, researchers
designed a textile-based wearable device for the unobtrusive recording of ECG, respiration
and accelerometric data and to assess the 3D sternal Seismo-CardioGram (SCG) in daily life.
Researchers also designed a portable and continuous Ballisto-CardioGram (BCG) monitor
that is wearable in the ear. The ear devices can reveal important information about cardiac
contractility and its regulation.
The Wearable Cardioverter Defibrillator (WCD) was introduced into clinical practice
in 2001, and indications for its use are currently expanding. The WCD represents an
alternative approach to prevent sudden arrhythmic death until either Implantable Cardioverter
Defibrillator (ICD) implantation is clearly indicated, or the arrhythmic risk is considered
significantly lower or even absent.
Kumar and Gandhi [2018] proposed a health monitoring system where a three-tier scalable
architecture has been developed. The system provides a real-time monitoring of patients data
where the study has been conducted by using Apache Mahout and the logistic regression
proved to be effective in predicting the presence and the absence of heart disease.
Wang et al., [2019] investigated a study on heart disease prediction and came up with
wearable device-based cardio-vascular monitoring systems. The wearable device software
is integrated with the cloud platform using a custom system-on-chip differentiating the tiers
utilized in the system. The system is capable of monitoring the electrocardiogram signals
which are placed on the patient’s body with the help of MicroController (MCU) and radio

29
frequency module. This enables the flexibility in acquiring the signals even from the remote
areas and then sends it to the storage for further processing. The cloud utilized performs the
scrutiny. The Integrated circuits installed provides a feasible solution for smaller inputs with
less electricity power and information transforming capacitance. The elimination of unwanted
noise can be done with the help of wavelet decomposition thereby improving the quality of the
signals received from the patient’s body. Different scenarios have been tested in the patients
while running and walking where the dataset is obtained from Massachusetts Institute of
Technology-Beth Israel (MIT-BIH) database thereby providing accuracy for prediction of
about 97%.
Wu et al., [2020] come with a heart disease prediction system by using the Electro-Cardio
Graph (ECG) signals where the system gets installed on a bodice. (i.e., Wearable). Bluetooth
Low Energy (BLE) is used to transmit, receive signals in real-time. The system comprises
BLE for synchronized exhibit, menu-driven linkage for PC display and smartphone. The
study comprises a system consisting of three important modules such as ECG measurement
(i.e., AD8232), ECG data processing and delivery module (i.e.,RFD77101), power controller.
Biopotential Analog Front End (AFE) circuits collect the data, which is processed by the
MCU and then transmitted to the application or a Personal Computer (PC) Graphical User
Interface (GUI) via BLE for storing and displaying the result. The ECG AFE, MCU, and the
BLE are powered by a 2.4 V 240 mAh rechargeable battery. Being a small, adaptable and
wearable device the device can be used even while on outside thereby making the possibility
to get charged using the solar heat. The components are embedded using a Printed Circuit
Board (PCB) which makes it gentle and to use it daily even while on outside. The device can
be able to operate and to monitor and to take signals from the patient’s body continuously.
Li et al., [2019] proposed a study of how far the old aged people have been compatible
with the wearable devices system. The study elaborates the significance of these wearable
devices in health and monitoring and early treatment. Aged people showed much passion
towards these smart wearable devices which reduced the huge burden to depend on the need
of physical human beings being present. These systems are age-friendly, light weight, small,
waterproof, compatible, gentle, lively and easy to wear at low cost which makes it flexible in
capturing the attention of the patients of the old-aged.
Constant et al., [2015] designed a cloud based computing wear glass which monitors the
heart rate continuously. Though several adjustments contribute in finding the solution there
are multiple difficulties such as disturbance, interference generated by the actions of human

30
parts also needs to be investigated.
Li et al., [2017] designed an IoT based heart disease monitoring system. Since the heart
attack people cannot be able to call the health services on the fainting stage. The proposed
system works by sending the important symptoms to a remote medical care facility even
located far away. The system continuously monitors and acquires signals from the patients
and reports it to the remote medical care in the real world based on the collected data and not
on feelings. The system is composed of three different layers such as: a sensing layer which
collects the data, the transport and the application layer. In addition to the location all these
data are recorded continuously. Several body conditions are taken into account to decide if
there is a need for emergency care, data such as ECG, Oxygen Saturation (SpO2), blood fat,
blood glucose, blood pressure, heart rate, pulse rate, in addition with the sufferers’ position
in order to facilitate timely reach under unforeseen circumstances. Since it is a wearable
system, it is cost effective thereby providing virtual reality where there is no necessity to
transmit information. The data collected can be sent by way of two-process. The data gets
transmitted through a sensor device connected to a short distance connected to a distant
server. There are various ways which work in the form of (1). Simultaneous uninterrupted
exchange of all data (2). Uninterrupted exchange across specific events (3) critical care
activated information exchange and (4) on-demand need. Studies have integrated with the
Data Stream Management System technologies and the decision making on the collected
information from the patient’s where the warning signals can be activated predominantly
before catching any distinct signs.
Koole et al., [2019] developed a m-Health telemonitoring application for arrhythmia, heart
failure and blood pressure for adults with Congenital Heart Disease(CHD). These diseases are
more prevalent among grown-ups leading the lives of the affected most miserable. Everyday
monitoring can be done by using the m-Health application which provides a facility thereby
facilitating the rise of cause of halting the fainting condition of a patient.
Priya et al., [2020b] presents a promising approach in the analysis of either structured or
unstructured datasets to produce very substantial pattern discoveries. By defining the generic
architecture for prediction model, we reviewed various papers involved in similar fields based
on suggested methodologies and their associated outcomes. The study discussed the pros
and cons of different prediction models using traditional ML and modern ML techniques.
Priya et al., [2020a] conducted a research study using state of the art technology of
Internet of Things (IoT), Cloud and AI-based analytics; thereby providing continuous look

31
up and on-time & advanced alerts not limited with the sufferers but also to the Medicare
workers. This facilitates information transmission to distant emergency care. The information
recorded live will estimate the heart rate, pulse, temperature and detect fall through wireless
devices and connect to cloud servers. The information thereby can be coagulated with the
older history of the patient and visualized critically using ML techniques in order to forecast
pre-judiciously.
Haq et al., [2018] Healthcare industry is one of the industries which facilitate a huge
growth of wearable devices and the use of the internet combined towards it.
Santana-Mancilla et al., [2020] comes up with an Internet of Health Things (IoHT)
system where the patient’s conditions can be monitored continuously. The data collected,
transmitted and received continuously uninterrupted even to a distant health care official at a
low-cost. This enables the physicians to be aware about the patient’s physical conditions which
facilitates diagnosis at an early stage. The significant improvement provides the healthcare
industry with the people to live long with good health and a better lifestyle compared to that
of the old healthcare styles. It enables old-aged people leading unaccompanied.
Lee et al., [2011] developed a system which collects different physical and psychological
data from tri-axial accelerometers and a heart-rate monitor equipment which helps to monitor
and to limit obesity. An end-user use case application has been utilized to differentiate
different physical actions using supervised machine learning algorithms which are Linear
Discriminant Classifier (LDC), Quadratic Discriminant Classifier (QDC), Uncorrelated
normal Density-based Classifier (UDC), Polynomial Classifier (POLYC), Logistic Classifier
(LOGLC), KNN, DT, Parzen Classifier, SVM and NB classifiers. This results in different
patterns and features being acquired and evaluated thereby providing an accuracy of about
99% and a sensitivity of 100%.
Al-Makhadmeh and Tolba [2019] developed automatic systems by the prognosing of
heart disease which provides improvisation before diagnosis. The information collected
from the patient’s heart by using IoT medical devices to be stored in the health care center
that can be processed by using the higher order Boltzmann deep belief neural network
(HOBDBNN). Deep learning methods utilized achieve the efficiency of about 99% accuracy
within a minimum span of time. The comparison of various existing systems for heart disease
prediction models is listed in Table 2.1.

32
 Table 2.1: Heart Disease Prediction using hybrid Machine Learning Classification Algorithms

System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
BNN: Accuracy
(74.5%),
Back-propagation
Sensitivity(89.7%),
Magneto - neural network
Ischemic Coronary Magneto - sensitivity (54.5%)
1 Tanawut [Link] (2008) Cardiography (BNN) Direct kernel
Disease (IHD) cardiograms DK-SOM:
(MCG) self-organizing map
Accuracy(80.4%),
(DK-SOM)
Sensitivity (86.2%),
33

Specificity(72.7%)
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
Back-Propagation
Neural Network
BPNN & BNN: (BPNN), Bayesian
Magneto- Accuracy (78.43%), Neural Network
Ischemic Coronary Magneto -
2 Yosawin [Link] (2010) Cardiography BNN: Sensitivity (BNN),
Disease (IHD) cardiograms
(MCG) (96.65%) SVM: Probabilistic neural
Specificity (86.36%) network (PNN),
34

Support vector
machine (SVM)
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
J4.8 Decision Trees
(DT): Sensitivity
(72.01%),
Specificity
(84.48%), Accuracy
J4.8 Decision Trees
(78.9%) Bagging
3 Mai [Link] (2011) Study Heart Disease — (DT), Bagging
Algorithm:
35

Algorithm
Sensitivity
(74.93%),
Specificity
(86.64%), Accuracy
(81.41%)
Coronary Heart K-NN:Accuracy K-Nearest neighbor
4 Mai [Link] (2012) Study —
Disease (CHD) (97.4%) Algorithm(K-NN)
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
SVM with
Sequential
Parthiban [Link] Cardio-Vascular AI based Neural
5 Study Minimization
(2012) Disease(CVD) Network System
Optimization
learning algorithm
Logistics,
Multi-Layer
36

Vijayarani et. al. Congenital Heart Perception and

6 Study — Accuracy
(2013) Disease (CHD) Sequential Minimal
Optimization
algorithms.
Feature Selection
Cardiovascular based Least Square
7 Divyva [Link] (2014) Study — Accuracy (85.59%)
Disease (CVD) Twin Support Vector
Machine (LSTSVM)
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
Coronary Artery
8 Ahmed et. al (2015) System Wearable Device Accuracy (85%) BayesNet, SVM, FT
Disease (CAD)
J48 algorithm,
Logistic model tree
Jaymin patel [Link] J4.8:
9 Study Heart Disease — algorithm and
(2016) Accuracy(56.76%)
Random Forest
algorithm
37

NaiveBayes based
Coronary Heart Decision Support
10 Nishant [Link] (2017) Study Accuracy (86.41%) on AdaBoost and
Disease (CHD) Model
Boosted Trees
K-NN, DT, SVM
Rong Tao [Link] Ischemic Coronary
11 Study Magnetocardiograms Accuracy (94.03%) with XGBoost
(2019) Disease (IHD)
Model
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
Logistic regression,
Coronary Heart Decision Support K-NN, ANN, SVM,
12 Amin [Link].(2018) System Sensitivity (100%)
Disease (CHD) Model NB, DT, and
random forests
ANN, SVM,
Coronary Heart Clinical Decision Logistic Regression,
13 Ashok [Link] (2018) Study Accuracy(85 %)
Diseas (CHD) Support SYstem Classification Tree,
38

Naïve Bayes
Adult Congenital
Machine Learning
14 Gerhard [Link] (2019) Study Heart Disease Accuracy(90.2 %) –
Model
(ACHD)
XGBoost,
Zhenzhen [Link] Cardiovascular k-NN,Random
15 Study — —-
(2019) Disease (CVD) forests, Logistic
Regression
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
Logistic Regression,
Multi-Layer
Cardiovascular Machine Learning Perceptron, Naïve
16 Khalid [Link] (2019) Study Accuracy (89%)
Disease (CVD) Model Bayes using
Majority Voting
Rule
Decision Tree,
39

Support Vector
Machine, Random
Cardiovascular Machine Learning
17 Hager [Link] (2020) System Accuracy (94.9%) Forest Classifier,
Disease (CVD) Model
and Logistic
Regression
Classifier
DT, NB, K-NN,
Mehrbaksh [Link] Cardiovascular Machine Learning
18 System BPNN:Accuracy(93%) ANN, RBF, and
(2020) Disease (CVD) Model
SVM
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
Principal
Component
Analysis (PCA),
[Link] [Link] Cardiovascular Machine Learning
19 Study Hybrid:Accuracy(97%) Self-Organizing
(2020) Disease (CVD) Model
Map, Fuzzy Support
Vector Machine
(Fuzzy SVM),
40

Ritu Aggrawal [Link] Cardiovascular Machine Learning LDA, RF, GBC, DT,
20 Study Accuracy(93%)
(2020) Disease (CVD) Model SVM, and KNN
k-nearest neighbor
Mamun Ali [Link] Cardiovascular Machine Learning (KNN), decision
21 Study Accuracy(100%)
(2021) Disease (CVD) Model tree (DT) and
random forests (RF)
Cardiovascular Machine Learning Modern Gradient
22 Priya [Link] (2021) Study Accuracy(86.8 %)
Disease (CVD) Model Boosting
Continued on next page
 Table 2.1 – Continued from previous page
System / Detection /
[Link]. Author / Date Application area Parameter Algorithm
Study Prediction Model
Hybrid approach for
feature selection &
Logistic Regression,
Bayesian-
Cardiovascular SVM, DT, XGB
23 Priya [Link] (2021) Study Optimization Accuracy(90.16%)
Disease (CVD)zz Classifier, KNN,
technique for
Random Forest,
Hyperparameter
Tuning
41

Cardiovascular Machine Learning MLP-Neural

24 Priya [Link] (2021) Study Accuracy(99%)
Disease (CVD) Model Network
 Table 2.2: Heart Disease Prediction using Unsupervised Machine Learning Algorithms

Detection /
[Link]. Author / Date System Application area Prediction Parameter Algorithm
Model
density based cluster,
Machine model based cluster,
Atul kumar pandey Cardiovascular
1 Study Learning Accuracy (82%) grid based cluster and
[Link] (2013) Disease (CVD)
Model partition based
clustering
Machine
Dr. Sandeep Kumar
2 Study HIV Learning Accuracy K-means Clustering
[Link] (2010)
Model
Principal Component
Machine
Cardiovascular Analysis (PCA) with
3 Negar [Link] (2014) Study Learning Accuracy(87%)
42

Disease (CVD) fuzzy c-means and

Model
K-means
Coronary Artery Machine
4 Stricker [Link] (2013) Study Disease(CAD) and Learning — PCA & KCA
Stroke Model
Machine
Cardiovascular CHI-PCA) with
5 Anna [Link] (2020) Study Learning Accuracy(98.7%)
Disease (CVD) random forests (RF)
Model
Machine PCA & Hybrid Genetic
Touhidul [Link] Cardiovascular
6 Study Learning Accuracy(94.06%) Algorithm with
(2020) Disease (CVD)
Model K-means CLustering
Accuracy: Support
k-nearest neighbor,
Machine Vector machine 83%
Cardiovascular decision tree, linear
7 Archana [Link] (2020) Study Learning Decision tree 79%
Disease (CVD) regression and support
Model Linear regression 78%
vector machine(SVM)
k-nearest neighbor 87%
 Table 2.3: Heart Disease Prediction using Wearable devices Pertaining to Rural Environment

Wireless
Author /
SL. No. System Application Area Detection Device Communication Algorithm
Date
Technology
Continuous heart Bluetooth, 4G LTE
Priyan et.
1 System rate (HR) Wearable Devices RF, wireless Logistic Regression
al.(2018)
monitoring communication
Chenyu Bluetooth, 4G LTE
Wearable Electro- cardiograph Wearable Patch of
2 Wang et. RF, wireless Threshold training algorithm
ECG monitoring system sensors
al.(2019) communication
43

Smart T-
Taiyang
shirt for Electro- cardiograph Wearable Smart T- Bluetooth low
3 Wu et al. Not specified
ECG monitoring system shirt energy (BLE)
(2019)
monitoring
Junde Li
Smart Health Monitoring Bluetooth low
4 [Link]. Wearable Devices Not specified
Wearables System energy (BLE)
(2019)
Continued on next page
 Table 2.3 – Continued from previous page
Wireless
Author /
SL. No. System Application Area Detection Device Communication Algorithm
Date
Technology
Nicholas Continuous heart
Pulse- Wearable Smart Wi-Fi, Internet,
5 Constant et rate (HR) Not specified
Glasses Glass BLE
al. (2015) monitoring
Sensors
Sensors wear in
Chao Li et. heart Heart disease Bluetooth, GSM,
6 wrist or carried by Not specified
al. (2019) disease mentoring and GPRS, ADSL
44

patients
mentoring
M. A. C. Wireless ECG,
Daily monitoring of
7 Koole et al. Hart Wacht pressure monitor Wi-Fi &Internet Not specified
CHD
(2019) and a weight scale
Sensor
Amin [Link]. based Disease Detection &
8 Wearable Devices Wi-Fi &Internet Not specified
(2019) Smart Monitoring
Wearables
Continued on next page
 Table 2.3 – Continued from previous page
Wireless
Author /
SL. No. System Application Area Detection Device Communication Algorithm
Date
Technology
Li et. al. Continuous heart
9 Sensors Wearable Devices Wi-Fi &Internet Not specified
(2017) rate (HR)
Continued on next page
45
 Table 2.3 – Continued from previous page
Wireless
Author /
SL. No. System Application Area Detection Device Communication Algorithm
Date
Technology
actions using supervised
machine learning algorithms
which are linear discriminant
classifier (LDC), quadratic
discriminant classifier (QDC),
uncorrelated normal
46

Chelsea density-based classifier (UDC),

Continuous heart
10 Dobbins et – Wearable Devices Wi-Fi &Internet polynomial classifier (POLYC),
rate (HR)
al. (2017) logistic classifier (LOGLC),
k-nearest neighbor (KNNC),
decision tree (TREEC), parzen
classifier (PARZENC), support
vector classifier (SVC) SVM
and Naïve Bayes classifier
(NAIVEBC).