Using machine learning to Predict Heart Disease

NIKHIL BORA, SREEDEVI GUTTA, AHMAD HADAEGH

Department of Computer Science and Information System

California State University San Marcos

333 Twin Oak valley Rd. San Marcos CA, 92009

UNITED STATES OF AMERICA

Abstract— Heart Disease has become one of the most leading cause of the death on the planet and it has

become most life-threatening disease. The early prediction of the heart disease will help in reducing death rate.

disease on the basis of features (such as cholesterol, blood pressure, age, sex, etc. which were extracted from

the datasets. In our research we used two separate datasets. The first heart disease dataset we used was collected

from very famous UCI machine learning repository which has 303 record instances with 14 different attributes

(13 features and one target) and the second dataset that we used was collected from Kaggle website which

contained 1190 patient’s record instances with 11 features and one target. This dataset is a combination of 5

popular datasets for heart disease. This study compares the accuracy of various machine learning techniques. In

our research, for the first dataset we got the highest accuracy of 92% by Support Vector Machine (SVM). And

for the second dataset, Random Forest gave us the highest accuracy of 94.12%. Then, we combined both the

datasets which we used in our research for which we got the highest accuracy of 93.31% using Random Forest.

muscular organ which placed directly behind and

slightly left of breastbone. Heart Disease causes

highest number of deaths globally, with

approximately around 17.9 million people died from

it every year which means around 31% of deaths are

from the heart disease as per the WHO (World

Health Organization). Heart disease are also called

as Cardiovascular Disease which are group of

complication of the blood vessels and heart which

include cerebrovascular disease, rheumatic heart

them.

People who are at risk of heart disease may have

symptoms like high blood pressure, obesity,

cholesterol, diabetes, age, etc. As there is recent

accuracy. The main objective of our project is to

enhance efficiency for predicting heart disease rate.

2. Related Work

Heart disease is the leading cause of death

nowadays. In the work of Umair Shafique et. al. [1],

the authors used data mining techniques, decision

tree, Naïve Bayes and Neural Network algorithms,

for which they got accuracy of 82% for Naïve Bayes

using decision tree.

In the work of Vikas Chaurasia et. al. [3], the

dataset the authors used in their work was

downloaded from UCI laboratory which has 14

different attributes out of which they only used 11

attributes to predict heart disease using Naïve Bayes

and Decision Tree machine learning algorithms.

WEKA tool was used in their work, using which

they achieved 82% accuracy for Naïve Bayes, and

for SVM. The lowest accuracy they got is using K-

Nearest Neighbors (KNN) of 68%. The dataset they

used in their work was unbalanced, resulting in a

need for applying sampling techniques. But they

directly applied machine learning algorithms

without filtering data in the dataset.

Malkari Bhargav et. al. [5], proposed to identify

heart disease using different ML techniques. They

collected dataset from UCI ML repository. Dataset

use to forecast heart disease using ML models. The

authors got highest accuracy score for KNN 83%

and lowest accuracy score for SVM of 65% and for

Naïve Bayes 80%. Apurb Rajdhan et. al. [7] used

ML algorithms like decision tree, logistic

machine learning algorithms. Baban Rindhe et. al.

[14] compared 3 data mining classifier techniques

such as Support vector classifier, Random forest

classifier and neural network among all these 3,

Support vector classifier proved to be effective as it

achieved the highest accuracy of 84% for predicting

heart disease.

H. Jayashree et. al. [15] did research on heart

disease diagnosis in which the authors used both

gradient boosting as well as classification

got 81% accuracy.

In comparison to the previous works, we used

combination of datasets in our research and got very

high accuracy.

The main goal of our proposed method is to predict

which will give the best accuracy for prediction of

heart disease.

The language that I used in this research is

understanding data much easier like matplotlib

which is the most used library for data visualization

as it enables us to generate histograms, plots and

charts for representing data. All these reasons make

python perfectly fit for machine learning projects. In

Table 1: Dataset 1: Heart disease dataset

collected from UCI machine learning repo.

Dataset (Comprehensive) collected from Kaggle

dependent variable is negatively correlated to some

of the factors like thalach, slope, cp and positively

correlated to some of the parameters like thal, ca,

etc. In simple words, heatmap is very useful for data

visualization as it tells us a lot about how these 14

attributes of datasets are correlated to each other.

Random Forest is the most famous and it is

considered as the best algorithm for machine

learning. It is a supervised learning algorithm. To

achieve more accurate and consistent prediction,

solve complicated classification problems. The

to be independent of the presence of any other

feature according to naïve Bayes theorem. The

formula is:

P(S|T) = P(T|S) * P(S) / P(T)

Here, T is the event to be predicted, S is the class

value for an event. This equation. will find out the

class in which the expected feature for

classification.

a line of hyperplanes. Data points which are closest

to the hyperplane or points of the data set that, if

deleted, would change the position of dividing the

hyperplane are known as support vectors. As a

result, they might be regarded as essential

components of the data set. The margin is the

distance between hyperplane and nearest data point

from either collection. The goal is to select the

hyperplane with the maximum possible margin

between it and any point in the training set

increasing the likelihood of a new data being

properly classified. SVM’s main objective is to find

hyperplane is actually dependent on the quantity of

input features. If input has two features in that case

the hyperplane will be a line and two-dimensional

simple and easy to implement and also has high

speed because of which it is preferred over the other

algorithms when it comes to solving classification

problems. The algorithm classifies whether or not

the patient has disease by taking the heart disease

dataset as an input. It takes input parameters like

age, sex, chol, etc and classify person with heart

learn, Matplotlib and Seaborn libraries were

used.

on our dataset. As we converted categorical

variables into numerical variables using

get_dummies() function of Pandas library. Both

our datasets contains some categorical variables

data by applying Standardization by using

StandardScalar() and fit_transform() functions

of scikit-learn library.

Y’s. X’s are features or input variables of our

datasets and Y’s are dependent or target

variables which are crucial for predicting

disease. Then using by the importing

model_selection function of the sklearn library,

we splitted our X’s and Y’s into train and test

accuracy.

We used precision, F1-score, recall and accuracy

evaluation metrics for evaluating our models. False

Positive (FP) is when a model incorrectly predicts a

positive outcome. False Negative (FN) is when a

model incorrectly predicts the negative outcome.

Table 3 shows that KNN gives us the best result

with training accuracy of 100% and testing accuracy

of 90.16%. SVM actually gave us the best testing

accuracy of 91.80% but we got less training

accuracy. Also, the other random forest and Logistic

regression models performed well.

Table 5: Results of ML models on Combined

are the results which we obtained after applying ML

models on Combined dataset:

As shown in table 5, for this combined dataset

Random Forest gave us the best testing accuracy of

93.31% and training 100%. XGBoost also gave us

91.64% of testing accuracy. Naïve Bayes and

Logistic Regression did not perform well in this

6. Conclusion and Future Work:

Forest for prediction of heart disease. We used two

datasets which are publicly available, one from the

UCI machine learning repository and another from

Kaggle. For the UCI heart dataset, we got the

highest test accuracy for SVM of 92%. When we

applied ML models on the second dataset(Kaggle)

we got the highest accuracy of 94.12% by using the

Random Forest algorithm, while the lowest

accuracy we got for Naïve Bayes of 79.83%. Then

we combined both the datasets which we used in our

to find the best parameters for the ML model. Also

also played a crucial role in giving us the best

accuracy. For Performance measurement, I used

precision, recall, TN, FP, FN, TP rate, evaluation

metrics. Based on the results which we got in our

research it indicates that applying machine learning

algorithms are really important to predict heart

disease in an early stage. Our system could assist

doctors in diagnosing heart disease. In the future, we

may investigate deep learning algorithms to see the

results.

Irfan Ul Mustafa on “Data Mining in

Sundareswarar on 'Machine Learning

diabetes mellitus prediction,” Neural Comput.

Heart Failure by Using Unstructured Data of

Cardiac Patients,” 2016 45th Int. Conf.

Parallel Process. Work., pp. 426–431, 2016

Diagnosis: A Machine Learning Interpretation

Approach " in 2019 (IJACSA) International

Journal of Advanced Computer Science and

Applications, Vol. 10, Issue 12, 2019