International Journal of Innovative Technology and Exploring Engineering (IJITEE)

ISSN: 2278-3075, Volume-9 Issue-5, March 2020

Heart Disease Prediction using Machine

Learning Techniques
Raparthi Yaswanth, Y. Md. Riyazuddin

 physicians are unable to predict the patient disease status in

Abstract: Recent advancement of technology allows the the earlier stage before they were getting in to the final stage. .
automation of things to be done using machine learning So in these cases if any machine learning techniques can be
techniques. These machine learning techniques can also be used used by physicians then there is a chance to predict the heart
for detecting or predicting the heart disease in the early phase.
The health care industry produces a huge amount of data which is disease status in the earlier phase which can save many lives
in unstructured manner that cannot be understood by a machine. with perfect treatment. The hospital administrator can store
Due to development of modern technology, health care industries patient’s details which is having huge data in database server.
also managing the data in a structured manner which can be Then the physicians can extract those data and make it useful
understood by machine learning technology. In this environment for heart disease prediction with machine learning algorithms.
if we use machine learning algorithms for prediction of heart
disease, then there is a chance to detect the heart disease status in
The supervised machine learning algorithms will work on
the early phase and to alert patient to get a better treatment to cure labeled dataset which contains independent and dependent
that disease. This paper implements seven supervised learning attributes. The machine learning algorithms can work on huge
algorithms which are KNN, Decision Tree, Naive Bayes, Logistic data or complex data easily by classifying them, so these kind
Regression, Random Forest, Support Vector Machine and Neural of techniques can reduce unknown diseases and
Networks for heart disease prediction. This paper generates
corresponding tasks. These machine learning techniques help
algorithm performance metrics like Accuracy, Precision, Recall,
F-score and ROC values for how the system was predicting physicians to predict heart disease so that they can diagnosis
accurately. In this paper among those seven algorithms, Neural them and check with proper treatment. The heart diseases can
Networks gave best accuracy as 92.30% and this system provides be predicted with the help of huge existing datasets and then
experimental results for how the model is accurate for heart by applying classifiers like KNN, Decision Tree, Naive
disease prediction. Bayes, Logistic Regression, Random Forest, Support Vector
Machine and Neural Networks. These models can give more
Keywords : Heart disease prediction, Decision Tree, Naive
Bayes, Logistic Regression, Random Forest, Support Vector accuracy with their medical datasets which are stored in
Machine, K-Nearest Neighbor, Neural Networks. repository which shows the information in a meaningful way
with machine learning techniques.
I. INTRODUCTION In this paper further sections include, section 2: Literature
survey, where we discuss the related work of this concept.
The rapidly increasing medical data generated from Section 3: Proposed methodology, we discuss about the
system model, database collection, pre-processing, feature
hospital information system signifies the era of machine
extractions and classification of our algorithm steps. Section
learning in the healthcare domain. Previously, hospitals used
4: Results, we compare seven machine learning classification
to store medical reports and patient data in paper-based
algorithms and show the result in graphs. Section 5:
record. By performing clinical test the physician can provide
Conclusion, Section 6: References.
diagnosis details which may take chances to wrong decision.
It leads to patient death due to lack of medical components.
II. LITERATURE SURVEY
But the physician who is highly experienced can tell about
decisions of diagnosis which is cost expensive and time Sellappan Palaniappan et al. described in [1], how the
consuming. In healthcare domain, every patient report will be Intelligent Heart Disease Prediction system was worked based
stored in storage server, so that hospital admin can manage on machine learning techniques such as Naive Bayes and
the data in structured way which can be understood by Decision Tree. But it is web-based application and they have
machine. The advantage of machine learning techniques is used .Net technology which is a dependent platform.
that the physicians can do analysis on their existing data Peter C. Austin et al. [2] described that the doctors can
which leads to lower cost and less diagnosis time. In the divide the patients into two categories like with disease and
modern society, the younger and older generation also getting without disease. Here to classify they have used tree based
heart attack diseases. Because of their food habits and life technique based on patient disease type, but this type of
style, risks of heart attack cases are being increased. The classification may have chances to get less accuracy. So that
for improvement of model accuracy they included machine
Revised Manuscript Received on March 02, 2020. learning techniques like support vector machine and random
* Correspondence Author forests.
Raparthi Yaswanth*, MTech (Data Science) in the department of CSE
at GITAM University, Hyderabad. E: mail: yaswanthraparthi@gmail.com
E: mail: yaswanthraparthi@gmail.com
Dr. Y. Md. Riyazuddin, Assistant Professor, Department of CSE at
GITAM University,Hyderabad. E: mail: riyazymd@gmail.com

Published By:
Retrieval Number: E2862039520/2020©BEIESP
Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.E2862.039520 1456 & Sciences Publication
 Heart Disease Prediction using Machine Learning Techniques

Shan Xu et al. [3] describes about Cardiovascular Disease them 164 were negatives, 139 were positives. The dataset is
prediction which is early detection and it is important to stored in CSV (Comma Separated Value) file format where
patient for treatment. This system focuses on providing more each row represents a single value.
accuracy with help of support vector machine and naive bayes
algorithms. This system consists of four parts like data
interface which has hospitals raw data and data preprocessing
for data integration and feature selection which is to retrieve
only useful attributes for getting more accuracy and
performing the classification of those attributes.
Mansoor et al. [4] said that this system is to develop and
validate prediction model by implementing multivariate
logistic regression, full and reduced random forest models. In
this method eleven variables are included in final model based
on backward elimination as well as the full random forest
model will have 32 variables and they were mitigated up to 17
variables by feature selection for getting more accuracy which
can be useful and accurate tool in clinical practice.

III. PROPOSED METHODOLOGY

A. System Model

Fig. 2.Heart Disease Dataset

C. Preprocessing
In this phase first we need to gather the training dataset
which is in CSV file format. For that we need to read the file
data with the help of pandas library for converting it to list
array. The input message which is given by the user should get
append to list array, because the machine cannot understand
the file format data. On completion of converting process, it
will remove null values if they are in training and testing data.
D. Standardization
In this, we need to apply standardization on given training
dataset for scaling of each column data for getting best
accuracy model. To perform scaling we use StandardScaler
python library class which contains fit transform function and
it takes input as dataset columns data which have higher
values.
E. Classification Techniques
Fig. 1.System Architecture
Naive Bayes
In Figure.1, the system needs to load heart disease training The Naive bayes machine learning algorithm is useful for
dataset which is accessed from UCI repository and select best categorizing documents and email spam filtering and this
attributes for standardization and then we will do algorithm is working based on Baye’s rule [6].
preprocessing if any row has null values or empty cells. Once
preprocessing and feature extraction is completed, then it will
split the dataset as training and testing and we choose machine
learning classifiers like SVM, KNN, NB, DT, RF, LR and NN (1)
for heart disease prediction, while the prediction process take
test data as input and it returns output as positive or negative Wherein:
with the help of training dataset. We also calculate the A is a class
accuracy measures by splitting of dataset as 70% training and B is a message
30% for testing. P (A) is a class probability
B. Data Collection
In this system the heart disease dataset is shown in Figure 2
which contains 0’s and 1’s where 0’s indicates to negative and
P (B) is the probability of a
1’s indicates to positive status. Dataset has imported from
message
UCI repository [5]. This dataset contains 303rows and among

Published By:
Retrieval Number: E2862039520/2020©BEIESP
Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.E2862.039520 1457 & Sciences Publication
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-9 Issue-5, March 2020

P (B|A) is conditional probability of the class for the given function predict () which takes input as test message.
message B
Decision Tree:
P (A|B) is conditional probability where message B
belongs to class A. A tree has a nodes and branches. Decision tree is a
In our system, for implementing Naive bayes algorithm we classifier in the form of tree structure. The tree has two types
use python library which is named as sklearn. naive_ bayes. of nodes which are decision nodes and leaf nodes. Here
Gaussian NB class. This algorithm have function like fit () decision nodes specify a choice or test, based on this it can
which will build the training model whose inputs are decide which direction it can go and leaf nodes indicates the
independent and dependent values of dataset and other classification of example or the value of example. The
function like predict() function which takes input as testing decision tree goes well with both classification and regression
values and then it can predict the heart disease status as problems. The classification means, having a group of data
positive or negative. and we are supposed to classify the data into predefined set of
classes. The representation for the classification and
K-Nearest Neighbor: regression tree (CART) is a binary tree. Each root node
The K-nearest neighbor classifier [7] is a simplest represents a single input variable (x) and a split on that
algorithm for prediction of any dataset with the help of variable. The leaf nodes of the tree contain an output variable
Euclidean distance. Here K means number of neighbor’s, so if (y) which is used to make a prediction. In decision tree the
we take k=1 then it gives very nearest neighbor as output. It greedy approach [9] is used to divide the input space called
works on voting based system that means while prediction it recursive binary splitting. In this procedure, different split
takes nearest neighbors votes. This output gets more votes points are trained and tested using a cost function. The split
that will be the predicted output value. with the best cost is selected and all input variables and all
D = (2) possible split points are evaluated and chosen in a greedy
manner [10] based on the cost function. For classification
Whereas: Gini cost function indicates how pure the nodes will become
D is a distance between x and y split points. The Gini index follows the below equation:
K is a nearest neighbor
x and y are independent attribute values G=1- (6)
In our system, for implementing K-nearest neighbor
If the decision tree applies the Information Gain then it can
algorithm we use python library from the sklearn API for
follow the below equation.
classification which is named as sklearn.neighbors.
KNeighborsClassifier class. This algorithm has function like E=- (7)
fit() which will build the training model whose inputs are Where pi denotes probability of class, E denotes entropy
independent and dependent values of dataset and another and G denotes Gini Index.In our system for implementing
function predict() which takes input as testing values then it Decision Tree algorithm we use python library which is
can predict the heart disease status as positive or negative. sklearn.tree.DecisionTreeClassifier class.
Support Vector Machine: Logistic Regression:
The support vector machine [8] is supervised machine The Logistic Regression [12] is a supervised machine
learning algorithm which is used for classification of learning algorithm where it can solve the classification
instances. It can separate the data linearly and for non-linear problem. It works same as linear regression. It can be used for
data it can use kernel functions. The SVM classifies two describing the data and explains about relationship between
classes with the help of hyper plane which has the largest one dependent variable and one variable. It is a regression
margin to separate the dataset in to classes. The margin model where it can build to predict the probability with given
between the two classes represents the longest distance input data which belongs to category “1”. It can classify when
between closets data points of those classes which are called a decision threshold came to picture which is very important
support vectors. to set the threshold value. The Threshold value can affect by
the majority of precision and recall values.
. +b 1 (3)
P= (8)
. +b -1 (4)
y ( . +b)-1 (5) In our system for implementing Logistic Regression
Wherein algorithm we use python library as sklearn. linear_ model.
b is a constant distance. Logistic Regression class.
w , u are vectors.
y is the output, where 1 is for positive samples and -1 for
negative samples.
In our system, for implementing support vector machine
algorithm we use python library which name was sklearn.
Svm. SVR Class. This algorithm has function like fit () which
is used to build the training model whose inputs are Random Forest:
independent and dependent values of dataset and other Random Forest [12] is a

Published By:
Retrieval Number: E2862039520/2020©BEIESP
Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.E2862.039520 1458 & Sciences Publication
 Heart Disease Prediction using Machine Learning Techniques

supervised machine learning technique that constructs predicts as positive.

multiple decision trees. The final decision is made based on True Negative (TP): In our system, given input is expected as
the outcome of the majority of the decision trees. The decision negative and also it predicts output as negative.
tree depends on a single tree which leads to suffer with low So, by using above metrics we can calculate Accuracy,
bias and high variances. To overcome this problem the Precision, Recall, F-score of our proposed classifiers.
random forest relies on many trees which introduces Accuracy: It correctly classifies the classes as True Positive
flexibility and converts high variance to low variance. The and True Negative over total number of classification.
random forest process can follow the below steps:
Step-1: Construct bootstrapped dataset: Observe the Accuracy = 100 (11)
randomness involved in constructing the bootstrapped data
set (random sampling with replacement). Precision: It retrieves fractions of messages which are
Step-2: Construct Decision tree using the bootstrapped relevant for users.
dataset. While constructing the decision tree, the candidates
Precision = (12)
for root node and for the rest of the nodes can be randomly
selected. Recall: It retrieves fractions of messages that are relevant to
Step-3: Repeat step-1 & step-2 to get more number of query.
decision trees. This method takes many base learners i.e. Recall = (13)
decision trees and aggregating the decision of the majority F-Score: It represents the mean of precision and recall.
population is called bagging. F-score = 2 (14)
The randomness involved in the training dataset makes the
Where P is a precision, R is a recall.
random forest classifier more accurate than unseen test
dataset. Thereby it is yielding in low variance compared to
IV. RESULTS
decision trees. In our system for implementing Random
Forest algorithm we use python library which is named as In our experimental results of heart disease prediction
sklearn. ensemble. RandomForestClassifier class. model, we compared seven different algorithms and
generated the performance metrics like accuracy, precision,
Neural Networks:
recall, f-score and roc. The results were shown in table 1. We
The neural network [11]technique is a collection of nodes designed and developed our implementation in the Windows
where every node is associated with their respective weights. operating system having 8 GB Ram and 1 TB HDD. Our
Nodes calculate sum of weights and it forwards to the Results concluded that Neural Networks algorithm gives best
activation function. This activation function generate or results compared to KNN, Decision Tree, Naive Bayes,
defines a particular output for a given node based on input. Logistic Regression, Random Forest, Support Vector
Y=f ( ) (9) Machine.
The activation function is of three types: linear function,
heaviside step function and sigmoid function. The linear
function is simple where it can calculate the sum of weights
and then it forwards linearly. The heaviside step function is a
condition based function. It provides two states only either 1
or 0. It can return 1 if the weighted sum (v) is greater than or
equal to threshold (a) value otherwise it returns 0. The
sigmoid function is a complexity function. It can depend on
input.
f(v) = (10)

In our system for implementing Neural Networks algorithm

we use python library named as sklearn. neural_
network.MLPClassifier class.
F. Evaluation Metrics
The metrics shows the performance of machine learning
classifiers and it calculates how many misclassifications it can
generate. The below metrics will show how it effects on our Fig.3. Classification Performance measurements
proposed system:
True Positive (TP): In our system, given input is expected
as positive and also it predicts output as positive.
False Positive (FP): The input is expected as negative but it
predicts as positive.
False Negative (FN): It expects input as negative but it

Published By:
Retrieval Number: E2862039520/2020©BEIESP
Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.E2862.039520 1459 & Sciences Publication
 International Journal of Innovative Technology and Exploring Engineering (IJITEE)
ISSN: 2278-3075, Volume-9 Issue-5, March 2020

Table 1: Results Table

Algorithm Accuracy Precision Recall F1-Score ROC

KNN 80.21 0.77 0.84 0.80 0.80

NB 57.14 0.53 0.90 0.67 0.58

DT 74.72 0.74 0.60 0.66 0.72

LR 81.31 0.77 0.82 0.79 0.81

RF 85.71 0.85 0.82 0.83 0.85

SVM 85.71 0.89 0.79 0.83 0.85
NN 92.30 0.97 0.87 0.92 0.92

V. CONCLUSION AUTHORS PROFILE

This paper deals with various machine learning techniques Raparthi Yaswanth is currently pursuing his
M.Tech (Data Science) in the department of
and comparing them to know which technique among them Computer Science Engineering at GITAM
gives best results for accurate prediction of heart diseases. University, Hyderabad. He has a degree of B.Tech
The algorithms and techniques involve ensemble methods and in Information Technology from Gitam University,
the combination of artificial Neural Networks provide better Visakhapatnam. His main area of interest include
Machine Learning and Deep Learning. E:mail:
accurate results. In future, it is expected to use the above yaswanthraparthi@gmail.com.
techniques for eliminating the existing drawbacks and
improve the survival rate for the wellbeing of mankind.
Dr.Y.Md.Riyazuddin completed his PhD in
CSE, and working as Assistant Professor in CSE
REFERENCES Dept at GITAM University,Hyderabad.His
Research Areas include Networks and
1. S. Palaniappan, R. Awang, “Intelligent Heart Disease Prediction
Security,Machine Learning, Neural Networks.He
System Using Data Mining Techniques,” IJCSNS International Journal
has published good number of Publications in
of Computer Science and Network Security, vol. 8, no. 8, August 2008.
Reputed Journals. He is Reviewer Member for
2. P. C. Austin, J. V. Tu, J. E. Ho, D. Levy, D. S. Lee, “Using Methods
various Reputed Journals.E: mail:
from Data Mining and Machine Learning Literature for Disease
riyazymd@gmail.com.
Classification and Prediction: a Case Study Examining Classification
of Heart Failure Subtypes,” Journal of Clinical Epidemiology 66
(2013) pp. 398-407, 2013.
3. S. Xu, Z. Zhang, D. Wang, J. Hu, X. Duan and T. Zhu, “Cardiovascular
Risk Prediction Method Based on CFS Subset Evaluation and Random
Forest Classification Framework,” 2017 IEEE 2nd International
Conference on Big Data Analysis, 2017.
4. H. M. Islam, Y. Elgendy, R. Segal, A. A. Bavry and J. Bian, “Risk
prediction model for in-hospital mortality in women with ST-elevation
myocardial infarction: A machine learning approach,” Journal of Heart
& Lung, pp. 1-7, 2017.
5. Heart disease Data Set from UCI Machine Learning
Repository,https://www.kaggle.com/ronitf/heart-disease-uci.
6. F. Peng, "Augmenting Naive Bayes Classifiers with Statistical
Language Models", Computer Science Department Faculty
Publication Series”, Paper 91, 2003.
7. M. Akhil, B. L. Deekshatulu, and P. Chandra, “Classification of
Heart Disease Using K- Nearest Neighbor and Genetic Algorithm,”
Procedia Technol., vol. 10, pp. 85–94, 2013.
8. J. S. Raikwal,Kanak Saxena "Performance Evaluation of SVM and
K-Nearest Neighbor Algorithm over Medical Data set",pp. 35-39 July
2012.
9. A. A. Pathan, M. Hasan, M. F. Ahmed, and D. M. Farid, “Educational
Data Mining: A Mining Model for Developing Students Programming
Skills,” 8th International Conference on Software, Knowledge,
Information Management and Applications (SKIMA), 2014.
10. M. Kamber and P. J. Han, Data Mining Concepts, and Techniques, 3rd
ed., 2012.
11. S.P. Rajamhoana, C. Akalya Devi "Analysis of Neural Networks Based
Heart Disease Prediction System",4-6 July 2018.
12. S. M. M. Hasan1, M. A. Mamun2, M. P. Uddin3 "Comparative
Analysis of Classification Approaches for Heart Disease Prediction",
8-9 Feb. 2018.

Published By:
Retrieval Number: E2862039520/2020©BEIESP
Blue Eyes Intelligence Engineering
DOI: 10.35940/ijitee.E2862.039520 1460 & Sciences Publication