SN Computer Science (2020) 1:345

https://doi.org/10.1007/s42979-020-00365-y

ORIGINAL RESEARCH

Heart Disease Prediction using Machine Learning Techniques

Devansh Shah1 · Samir Patel1 · Santosh Kumar Bharti1

Received: 27 September 2020 / Accepted: 2 October 2020 / Published online: 16 October 2020
© Springer Nature Singapore Pte Ltd 2020

Abstract
Heart disease, alternatively known as cardiovascular disease, encases various conditions that impact the heart and is the
primary basis of death worldwide over the span of the past few decades. It associates many risk factors in heart disease and a
need of the time to get accurate, reliable, and sensible approaches to make an early diagnosis to achieve prompt management
of the disease. Data mining is a commonly used technique for processing enormous data in the healthcare domain. Research-
ers apply several data mining and machine learning techniques to analyse huge complex medical data, helping healthcare
professionals to predict heart disease. This research paper presents various attributes related to heart disease, and the model
on basis of supervised learning algorithms as Naïve Bayes, decision tree, K-nearest neighbor, and random forest algorithm.
It uses the existing dataset from the Cleveland database of UCI repository of heart disease patients. The dataset comprises
303 instances and 76 attributes. Of these 76 attributes, only 14 attributes are considered for testing, important to substantiate
the performance of different algorithms. This research paper aims to envision the probability of developing heart disease in
the patients. The results portray that the highest accuracy score is achieved with K-nearest neighbor.

Keywords  Heart disease prediction · Data mining · Decision tree · Naïve Bayes · K-NN · Random forest · Machine learning

Introduction hypertension, high blood cholesterol, and pre-existing heart

conditions are predisposing factors for heart disease. The
Over the last decade, heart disease or cardiovascular remains efficient and accurate and early medical diagnosis of heart
the primary basis of death worldwide. An estimate by the disease plays a crucial role in taking preventive measures to
World Health Organization, that over 17.9 million deaths prevent death.
occur every year worldwide because of cardiovascular Data mining refers to the extraction of required informa-
disease, and of these deaths, 80% are because of coronary tion from huge datasets in various fields such as the medical
artery disease and cerebral stroke [1]. The vast number of field, business field, and educational field. Machine learn-
deaths is common amongst low and middle-income coun- ing is one of the most rapidly evolving domains of artificial
tries [2]. Many predisposing factors such as personal and intelligence. These algorithms can analyse huge data from
professional habits and genetic predisposition accounts for various fields, one such important field is the medical field.
heart disease. Various habitual risk factors such as smok- It is a substitute to routine prediction modeling approach
ing, overuse of alcohol and caffeine, stress, and physical using a computer to gain an understanding of complex and
inactivity along with other physiological factors like obesity, non-linear interactions among different factors by reducing
the errors in predicted and factual outcomes [3]. Data mining
is exploring huge datasets to extract hidden crucial decision-
This article is part of the topical collection “Advances in making information from a collection of a past repository
Computational Approaches for Artificial Intelligence, Image for future analysis. The medical field comprises tremendous
Processing, IoT and Cloud Applications” guest edited by Bhanu
Prakash K N and M. Shivakumar”. data of patients. These data need mining by various machine
learning algorithms. Healthcare professionals do analysis of
* Devansh Shah these data to achieve effective diagnostic decision by health-
[email protected] care professionals. Medical data mining using classification
1
Computer Engineering Department, School of Technology,
algorithms provides clinical aid through analysis. It tests the
Pandit Deendayal Petroleum University, Raisan,
Gandhinagar 382007, India

SN Computer Science
Vol.:(0123456789)
345 
Page 2 of 6 SN Computer Science (2020) 1:345

classification algorithms to predict heart disease in patients training and test dataset. Training dataset trains our model
[4]. while testing dataset functions as new data to get accuracy
Data mining is the process of extracting valuable data of the model. The dataset exists with models and its output.
and information from huge databases. Various data mining The classification and regression are its example.
techniques such as regression, clustering, association rule
and classification techniques like Naïve Bayes, decision tree, Unsupervised Learning
random forest and K-nearest neighbor are used to classify
various heart disease attributes in predicting heart disease. Data used to train are not classified or labelled in the data-
A comparative analysis of the classification techniques is set. Aim is to find hidden patterns in the data. The model is
used [5]. In this research, I have taken dataset from the UCI trained to develop patterns. It can easily predict hidden pat-
repository. The classification model is developed using clas- terns for any new input dataset, but upon exploring data, it
sification algorithms for prediction of heart disease. In this draws conclusion from datasets to describe hidden patterns.
research, a discussion of algorithms used for heart disease In this technique, no responses in the dataset are seen. The
prediction, comparison among the existing systems is made. clustering method is an example of an unsupervised learn-
It also mentions further research and advancement possibili- ing technique.
ties in the paper.
Reinforcement Learning

It does not use labelled dataset nor the results are associated
Background
with data, thus model learns from the experience. In this
technique, the model improves its presentation based on its
Heart disease affects millions of people, and it remains the
association with environment and figures out how to discuss
chief cause of death in the world. Medical diagnosis should
its faults and to get the right outcome through assessment
be proficient, reliable, and aided with computer techniques
and testing various prospects.
to reduce the effective cost for diagnostic tests. Data mining
Classification algorithms are commonly used supervised
is a software technology that helps computers to build and
learning techniques to define probability of heart disease
classify various attributes. This research paper uses classifi-
occurrence.
cation techniques to predict heart disease. This section gives
a portrayal of the related subjects like machine learning and
its methods with brief descriptions, data pre-processing, Classification Machine Learning Techniques
evaluation measurements and description of the dataset used
in this research. The classification task is used for prediction of subsequent
cases dependent on past information. Many data mining
techniques as Naïve Bayes, neural network, decision tree
Machine Learning have been applied by researchers to have a precision diag-
nosis in heart disease. The accuracy given by different tech-
Machine learning is an emerging subdivision of artificial niques varies with number of attributes. This research pro-
intelligence. Its primary focus is to design systems, allow vides diagnostic accuracy score for improvement of better
them to learn and make predictions based on the experi- health results. We have used WEKA tool in this research
ence. It trains machine learning algorithms using a training for pre-processing the dataset, which is in ARFF format
dataset to create a model. The model uses the new input data (attribute-relation file format). Only 14 attributes of all 76
to predict heart disease. Using machine learning, it detects different attributes have been considered for analysis to get
hidden patterns in the input dataset to build models. It makes precise results. By comparison and analysis using different
accurate predictions for new datasets. The dataset is cleaned algorithms with WEKA tool heart disease can be predicted
and missing values are filled. The model uses the new input and treated early and prompt [5].
data to predict heart disease and then tested for accuracy.
Machine learning techniques are classified as:
Approach Methodology
Supervised Learning
This research aims to foresee the odds of having heart dis-
ease as probable cause of computerized prediction of heart
The model is trained on a dataset that is labelled. It has
disease that is helpful in the medical field for clinicians and
input data and its outcomes. Data are classified and split into

SN Computer Science
SN Computer Science (2020) 1:345 Page 3 of 6  345

patients [5]. To accomplish the aim, we have discussed the

use of various machine learning algorithms on the data set
and dataset analysis is mentioned in this research paper. This
paper additionally depicts which attributes contribute more
than the others to anticipation of higher precision. This may
spare the expense of different trials of a patient, as all the
attributes may not contribute such a substantial amount to
expect the outcome [5].

Data Source

For this study, I have used dataset from UCI Machine learn-
ing repository. It comprises a real dataset of 300 examples
of data with 14 various attributes (13 predictors; 1 class) like
blood pressure, type of chest pain, electrocardiogram result,
etc. (Table 1). In this research, we have used four algorithms
to get reasons for heart disease and create a model with the
maximum possible accuracy.

Data Pre‑processing

The real-life information contains large numbers with miss-

ing and noisy data. These data are pre-processed to over-
come such issues and make predictions vigorously. Figure 1
explains the sequential chart of our proposed model.
Cleaning the collected data usually has noise and miss-
ing values. To get an accurate and effective result, thes data
need to be cleaned in terms of noise and missing values are
to be filled up.
Transformation it changes the format of the data from one
form to another to make it more comprehensible. It involves
smoothing, normalization, and aggregation tasks. Fig. 1  Sequential chart of proposed model

Table 1  Attributes and details of dataset of heart disease

Sr. no. Attribute Representative icon Details

1 Age Age Patients age, in years

2 Sex Sex 0 = female; 1 = male
3 Chest pain Cp 4 types of chest pain (1—typical angina; 2—atypical angina; 3—non-anginal
pain; 4—asymptomatic)
4 Rest blood pressure Trestbps Resting systolic blood pressure (in mm Hg on admission to the hospital)
5 Serum cholesterol Chol Serum cholesterol in mg/dl
6 Fasting blood sugar Fbs Fasting blood sugar > 120 mg/dl (0—false; 1—true)
7 Rest electrocardiograph Restecg 0—normal; 1—having ST-T wave abnormality; 2—left ventricular hypertrophy
8 MaxHeart rate Thalch Maximum heart rate achieved
9 Exercise-induced angina Exang Exercise-induced angina (0—no; 1—yes)
10 ST depression Oldpeak ST depression induced by exercise relative to rest
11 Slope Slope slope of the peak exercise ST segment (1—upsloping; 2—flat; 3—down sloping)
12 No. of vessels Ca No. of major vessels (0–3) colored by fluoroscopy
13 Thalassemia Thal Defect types; 3—normal; 6—fixed defect; 7—reversible defect
14 Num(class attribute) Class diagnosis of heart disease status (0—nil risk; 1—low risk; 2—potential risk; 3—
high risk; 4—very high risk)

SN Computer Science
345 
Page 4 of 6 SN Computer Science (2020) 1:345

Integration the data may not be acquired from a single This algorithm splits the data into two or more analogous
source but varied sources, and it has to be integrated before sets based on the most important indicators. The entropy of
processing. each attribute is calculated and then the data are divided,
Reduction the data gained are complex and require to be with predictors having maximum information gain or mini-
formatted to achieve effective results. mum entropy:
The data are then classified and split into training data c
∑
set and test data set which is run on various algorithms to Entropy(S) = −Pi log2 Pi,
achieve accuracy score results. i=1

Algorithms Used ∑ |Sv|

Gain (S, A) = Entropy(S) − Entropy (Sv).
v∈ Values(A)
|S|
Naïve Bayes’ Classifier
The results obtained are easier to read and interpret [3].
Naïve Bayes classifier is a supervised algorithm. It is a sim- This algorithm has higher accuracy in comparison to other
ple classification technique using Bayes theorem. It assumes algorithms as it analyzes the dataset in the tree-like graph.
strong (Naive) independence among attributes. Bayes theo- However, the data may be over classified and only one attrib-
rem is a mathematical concept to get the probability. The ute is tested at a time for decision-making.
predictors are neither related to each other nor have correla- An accuracy of 71.43% has been achieved by the decision
tion to one another. All the attributes independently con- tree by Chauhan et al. [9], whereas the accuracy obtained
tribute to the probability to maximize it. It is able to work was very poor about 42.8954% in [10].
with Naïve Bayes model and does not use Bayesian methods.
Many complex real-world situations use Naive Bayes clas- K‑Nearest Neighbor (K‑NN)
sifiers [6]:
The K-nearest neighbors algorithm is a supervised classifi-
P(Y∕X) × P(X)
P(X∕Y) = , cation algorithm method. It classifies objects dependant on
P(Y) nearest neighbor. It is a type of instance-based learning. The
P(X/Y) is the posterior probability, P(X) is the class prior calculation of distance of an attribute from its neighbors
probability, P(Y) is the predictor prior probability, P(Y/X) is is measured using Euclidean distance [3]. It uses a group
the likelihood, probability of predictor. of named points and uses them on how to mark another
Naïve Bayes is a simple, easy to implement, and efficient point. The data are clustered based on similarity amongst
classification algorithm that handles non-linear, complicated them, and is possible to fill the missing values of data using
data. However, there is a loss of accuracy as it is based on K-NN. Once the missing values are filled, various predic-
assumption and class conditional independence. tion techniques apply to the data set. It is possible to gain
An accuracy of 84.1584% has been achieved in Naïve better accuracy by utilizing various combinations of these
Bayes with 10 most important predictors chosen using SVM- algorithms.
RFE [7]. While, the accuracy of 83.49% has been achieved K-NN algorithm is simple to carry out without creat-
using all 13 attributes of Cleveland dataset [8]. ing a model or making other assumptions. This algorithm
is versatile and is used for classification, regression, and
search. Even though K-NN is the simplest algorithm, noisy
Decision Tree and irrelevant features affect its accuracy. In a study by
Pouriyeh et al., 83.16% accuracy was achieved with value
Decision tree is a classification algorithm that works on cat- K = 9 [8].
egorical as well as numerical data. Decision tree is used
for creating tree-like structures. Decision tree is simple and
Random Forest Algorithm
widely used to handle medical dataset. It is easy to imple-
ment and analyse the data in tree-shaped graph. The decision
Random forest algorithm is a supervised classification
tree model makes analysis based on three nodes.
algorithmic technique. In this algorithm, several trees cre-
ate a forest. Each individual tree in random forest lets out a
• Root node: main node, based on this all other nodes func-
class expectation and the class with most votes turns into a
tions.
model’s forecast. In the random forest classifier, the more
• Interior node: handles various attributes.
number of trees give higher accuracy. The three common
• Leaf node: represent the result of each test.
methodologies are:

SN Computer Science
SN Computer Science (2020) 1:345 Page 5 of 6  345

Table 2  Percentage accuracy Naïve Bayes K-nearest neighbor Decision tree Random forest
results of classification
techniques Accuracy
 Testing set 88.15789474 78.94736842 (K = 2) 73.68421052631578 84.21052631578947
(parameter = entropy) (parame-
ter = entropy)
 Training set 81.05726872 90.78947368 (K = 7) 80.26315789473685 82.89473684210526
(parameter = gini) (parameter = gini)

Table 3  Accuracy of heart disease prediction with different tech-

niques
Authors Technique Accuracy

Otoom et al. [12] Naïve Bayes 84.5%

SVM 84.5%
Functional trees 84.5%
Vembandasamy et al. [13] Naïve Bayes 86.419%
Chaurasia et al. [14] J48 84.35%
Bagging 85.03%
SVM 94.60%
Parthiban et al. [15] Naïve Bayes 74%
Seema et al. [16] Naïve Bayes 95.556%
Decision tree Highest accuracy
Fig. 2  Comparative result of classification techniques obtained by
SVM
SVM
Kumar Dwivedi [17] Naïve Bayes 83%
• Forest RI (random input choice); Classification tree 77%
• Forest RC (random blend); K-NN 80%
• Combination of forest RI and forest RC. Logistic regression 85%
SVM 82%
It is used for classification as well as regression task, ANN 84%
but can do well with classification task, and can overcome The proposed model Naïve Bayes 88.157%
missing values. Besides, being slow to obtain predictions K-NN 90.789%
as it requires large data sets and more trees, results are Decision tree 80.263%
unaccountable. Random forest 86.84%
Random forest algorithm has obtained an accuracy of
91.6% with Cleveland dataset in [11]. Using People’s data-
set, an accuracy of 97% was achieved. K-nearest neighbor, and random forest are applied to get
accuracy score. The accuracy score results of different clas-
sification techniques were noted using Python Programming
for training and test data sets. Percentage accuracy scores
Results and Analysis are depicted in Table 2 and Fig. 2 for different algorithms.
Comparison of accuracy score of heart disease prediction in
Aim of this research is to predict whether or not a patient proposed model with different authors is given in Table 3.
will develop heart disease. This research was done on super-
vised machine learning classification techniques using Naïve
Bayes, decision tree, random forest, and K-nearest neigh- Conclusion
bor on UCI repository. Various experiments using different
classifier algorithms were conducted through the WEKA The overall aim is to define various data mining techniques
tool. Research was performed on 8th generation Intel Corei7 useful in effective heart disease prediction. Efficient and
having an 8750H processor up to 4.1 GHz CPU and 16 GB accurate prediction with a lesser number of attributes and
ram. Dataset was classified and split into a training set and tests is our goal. In this study, I consider only 14 essential
a test set. Pre-processing of the data is done and supervised attributes. I applied four data mining classification tech-
classification techniques such as Naïve Bayes, decision tree, niques, K-nearest neighbor, Naive Bayes, decision tree,

SN Computer Science
345 
Page 6 of 6 SN Computer Science (2020) 1:345

and random forest. The data were pre-processed and then 7. Pahwa K, Kumar R. Prediction of heart disease using hybrid
used in the model. K-nearest neighbor, Naïve Bayes, and technique for selecting features. In: 2017 4th IEEE Uttar Pradesh
section international conference on electrical, computer and elec-
random forest are the algorithms showing the best results tronics (UPCON). IEEE. p. 500–504.
in this model. I found the accuracy after implementing four 8. Pouriyeh S, Vahid S, Sannino G, De Pietro G, Arabnia H, Gutier-
algorithms to be highest in K-nearest neighbors (k = 7). rez J. A comprehensive investigation and comparison of machine
We can further expand this research incorporating other learning techniques in the domain of heart disease. In: 2017 IEEE
symposium on computers and communications (ISCC). IEEE. p.
data mining techniques such as time series, clustering and 204–207.
association rules, support vector machine, and genetic 9. Chauhan R, Bajaj P, Choudhary K, Gigras Y. Framework to pre-
algorithm. Considering the limitations of this study, there dict health diseases using attribute selection mechanism. In: 2015
is a need to implement more complex and combination 2nd international conference on computing for sustainable global
development (INDIACom). IEEE. p. 1880–84.
of models to get higher accuracy for early prediction of 10. Bouali H, Akaichi J. Comparative study of different classifica-
heart disease. tion techniques: heart disease use case. In: 2014 13th interna-
tional conference on machine learning and applications. IEEE. p.
482–86.
Author contribution  All authors have equal contribution to this study 11. Xu S, Zhang Z, Wang D, Hu J, Duan X, Zhu T. Cardiovascular
and all authors have read and approved the final manuscript. risk prediction method based on CFS subset evaluation and ran-
dom forest classification framework. In: 2017 IEEE 2nd interna-
tional conference on big data analysis (ICBDA). IEEE. p. 228–32.
Compliance with Ethical Standards  12. Otoom AF, Abdallah EE, Kilani Y, Kefaye A, Ashour M. Effective
diagnosis and monitoring of heart disease. Int J Softw Eng Appl.
Conflict of interest  The authors declare that they have no competing 2015;9(1):143–56.
interest. 13. Vembandasamy K, Sasipriya R, Deepa E. Heart diseases detec-
tion using Naive Bayes algorithm. Int J Innov Sci Eng Technol.
2015;2(9):441–4.
14. Chaurasia V, Pal S. Data mining approach to detect heart diseases.
References Int J Adv Comput Sci Inf Technol (IJACSIT). 2014;2:56–66.
15. Parthiban G, Srivatsa SK. Applying machine learning methods in
1. Seckeler MD, Hoke TR. The worldwide epidemiology of acute diagnosing heart disease for diabetic patients. Int J Appl Inf Syst
rheumatic fever and rheumatic heart disease. Clin Epidemiol. (IJAIS). 2012;3(7):25–30.
2011;3:67. 16. Deepika K, Seema S. Predictive analytics to prevent and con-
2. Gaziano TA, Bitton A, Anand S, Abrahams-Gessel S, Murphy A. trol chronic diseases. In: 2016 2nd international conference on
Growing epidemic of coronary heart disease in low-and middle- applied and theoretical computing and communication technology
income countries. Curr Probl Cardiol. 2010;35(2):72–115. (iCATccT). IEEE. p. 381–86.
3. Weng SF, Reps J, Kai J, Garibaldi JM, Qureshi N. Can machine- 17. Dwivedi AK. Performance evaluation of different machine learn-
learning improve cardiovascular risk prediction using routine ing techniques for prediction of heart disease. Neural Comput
clinical data? PLoS ONE. 2017;12(4):e0174944. Appl. 2018;29(10):685–693.
4. Ramalingam VV, Dandapath A, Raja MK. Heart disease predic-
tion using machine learning techniques: a survey. Int J Eng Tech- Publisher’s Note Springer Nature remains neutral with regard to
nol. 2018;7(2.8):684–7. jurisdictional claims in published maps and institutional affiliations.
5. Patel J, TejalUpadhyay D, Patel S. Heart disease prediction
using machine learning and data mining technique. Heart Dis.
2015;7(1):129–37.
6. Fatima M, Pasha M. Survey of machine learning algorithms for
disease diagnostic. J Intell Learn Syst Appl. 2017;9:1–16. https​
://doi.org/10.4236/jilsa​.2017.91001​.

SN Computer Science