IEEE Conference ID: 62700 13th – 15th Nov.

2024
Proceedings of the 2024 4th International Conference on Technological Advancements in Computational Sciences

Heart Disease Prediction Using Machine Learning:

A Data-Driven Approach
2024 4th International Conference on Technological Advancements in Computational Sciences (ICTACS) | 979-8-3503-8749-0/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICTACS62700.2024.10840998

Abboskhujaev Akhrorkhuja1, Danish Ather2, Rahul Chauhan3, Kireet Joshi4, Gurinder Singh5, Naina Chaudhary6
Amity University Tashkent, Tashkent, Uzbekistan
1,2,6

Computer Science & Engineering, Graphic Era Hill University, Dehradun, Uttarakhand, India
3

4
Computer Science & Engineering, Graphic Era Deemed to be University, Dehradun, Uttarakhand, India
5
Amity University Noida, Noida, Uttar Pradesh, India
E-mail: [email protected], [email protected],
1

3
[email protected], [email protected]

Abstract—Heart sickness remains a main purpose of diagnosis of heart disease can significantly improve patient
mortality worldwide, accounting for a significant percentage outcomes, allowing timely interventions and reducing the
of worldwide deaths. Early detection and correct diagnosis burden on healthcare systems.
are important in reducing its impact and enhancing affected Heart disease risk factors have been traditionally
person effects. This look at proposes a machine learning-
identify using clinical algorithms including the Framingham
based totally approach for heart sickness prediction, utilising
a dataset of scientific fitness parameters along with age,
Risk Score and other statistical models to help healthcare
gender, levels of cholesterol, and blood pressure. The version providers find those individuals at highest risk. But these
is evolved the usage of a mixture of algorithms, inclusive of approaches often use reduced data and ignore complex
logistic regression, random woodland, and aid vector machines interactions of multiple risk factors. Big data and machine
(SVM), to are expecting the probability of heart sickness. learning are a new frontier in predictive modelling, capturing
Feature choice strategies are implemented to discover the overarching patterns in vast amounts of data which have
most crucial parameters influencing heart disorder danger. represented as a very attractive alternative to heart disease
The dataset is break up into training and testing units, and the prediction models. With machine learning techniques,
models are evaluated primarily based on accuracy, precision, patient data will be analyzed and can now highlight the
bear in mind, and F1-score. Our experimental outcomes
major risk factors and even determine what is the likelihood
display that the random forest version performed the very
best performance with an accuracy of 85%, outperforming
of heart disease more accurately than traditional methods.
different fashions. This technique demonstrates the ability of Many researchers have applied machine learning model
machine gaining knowledge of in helping early prognosis and of the prediction of heart disease in recent years such kind
customized treatment making plans for sufferers susceptible of datasets are related to various clinical and demographic
to coronary heart sickness. The proposed method can be information. Predictive models with different machine
incorporated into healthcare structures to enhance predictive learning techniques like logistic regression, decision trees,
skills and facilitate proactive healthcare interventions. Future random forest and support vector machines (SVM) have
work will discover the inclusion of more various datasets to been used. These models win over traditional statistic
enhance the model’s generalizability.
models, especially in case of non-linear relationships
Keywords: Heart disease, Machine learning, Healthcare,
prediction, Data science
among variables. While models have improved in the task
of detecting recurrences, issues continue to present itself
I. INTRODUCTION when trying to generalize these predictions on different
According to the World Health Organization (WHO), patient groups.
cardiovascular diseases (CVD), commonly known as heart This studies ambitions to develop a robust and correct
diseases, cause approximately 17.9 million deaths each year machine gaining knowledge of version for predicting heart
and remain the leading cause of death worldwide. CVD disease, leveraging a dataset of medical parameters. By
refers to various heart conditions, including coronary artery employing multiple machine mastering algorithms, which
disease, heart attacks, and heart failure, which are the most include logistic regression, random forests, and SVM, this
common forms. Both modifiable (such as diet, smoking, observe seeks to discover the most influential elements
and physical inactivity) and non-modifiable (such as age contributing to heart sickness and evaluate the performance
and genetics) risk factors contribute to the high incidence of of every model in terms of accuracy, precision, keep in mind,
heart disease. Angina pectoris is a condition often associated and F1-rating. In doing so, the take a look at highlights the
with coronary artery disease. Accurate prediction and early capacity of gadget studying to revolutionize heart ailment

Copyright © IEEE–2024 ISBN: 979-8-3503-8749-0 23

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 04,2025 at 17:13:22 UTC from IEEE Xplore. Restrictions apply.
 IEEE Conference ID: 62700 13th – 15th Nov. 2024
Proceedings of the 2024 4th International Conference on Technological Advancements in Computational Sciences

prediction, facilitating early detection and taking into random forest model performed better than decision trees
consideration extra focused interventions. with an accuracy of 85% in a comparative study [5].
Furthermore, the model that is proposed can also be Other algorithms like support vector machines (SVM) are
integrated in healthcare systems to provide clinicians with a also widely used to provide high accuracy and precision
robust tool for determining heart disease risk and initiating in classifying heart disease tasks [3]. In another work,
interventions prior to the development of untoward hybrid SVM Particle Swarm Optimization (PSO) was
conditions. The rest of the paper is organized as follows: combined with traditional classification technique for
Section II provides a background on heart disease prediction higher performance which resulted in accuracy level of
using machine learning. In Section III, we discuss the 87% [6]–[9]. Recent improvements in deep mastering have
causes of heart disease and what key risk factors to look also made strides in coronary heart sickness prediction.
for. In the fourth section, we explain the methodologies Convolutional neural networks (CNNs) were carried out to
followed in building a prediction model which includes data categorise medical pictures associated with cardiovascular
preprocessing steps, feature selection process and training health, yielding high prediction accuracy while utilized in
an actual model. Section V presents experimental results, mixture with scientific information [10], [11]. However, the
and Section VI draws conclusions and proposes directions venture with deep getting to know models lies in their want
for future work. for big datasets and excessive computational electricity,
limiting their huge use in a few healthcare settings.
II. RELATED WORK
A not unusual difficulty confronted via most gadget
In recent years, the interest in predicting heart disease gaining knowledge of models is overfitting, mainly whilst
using machine learning algorithms has increased rapidly. working with small datasets. To mitigate this, techniques
Example different models and methodologies have been together with move-validation and regularization had been
tried to address this problem from the typical statistical extensively employed [12], [13]. Despite the high accuracy
model to advanced machine learning-frameworks. suggested in diverse studies, generalizability throughout
Various studies supports that machine learning can be specific populations stays a mission, and the inclusion
used to increase the accuracy in heart disease prediction. of numerous datasets is important for improving model
For instance, one of the standard models: logistic regression robustness.
yields as baseline performance for prediction (especially
Table I: Comparison of Machine Learning Models for Heart Disease
along with feature selection methods). For example, in one Prediction
study there was 79% classification accuracy by logistic
Model Accuracy Advantages Disadvantages
regression but model performance was proscribed due to
Logistic Simple,
assumptions of linearity. [1]–[4]. Regression
79% [1]
interpretable
Assumes linearity
A. Causes of Heart Disease Captures
Heart disease is influenced by a combination of genetic Decision Tree 82% [5] non-linear Prone to overfitting
relationships
and environmental factors. Fig. 1 illustrates the main causes
Robust, High computational
contributing to heart disease. Random Forest 85% [5]
reducesoverfitting cost
Support Vector Effective in
Hyperparameter
Machine ( 83% [14] highdimensional
tuning needed
SVM ) spaces
Combines SVM
SVM + PSO 87% [6] Increased complexity
with optimization
Convolutional
Neural High accuracy Requires large
90% [10]
Network with large datasets datasets
(CNN)

III. METHODOLOGY
In this section, we describe the methodology used for
heart disease prediction. Figure 2 provides an overview
Fig. 1: Causes of Heart Disease of the methodology, including data preprocessing, model
As shown in Fig. 1 First, smoking, an unhealthy diet training, and evaluation.
and low physical activity are major lifestyle factors. On
top of that, things such as high blood pressure, cholesterol
and diabetes make the risk even higher. Decision trees
and random forest machines demonstrated the greatest
improvement, as they are able to capture non-linear
relationships between risk factors and heart disease. The Fig. 2: Overview of the Methodology for Heart Disease Prediction

24 Copyright © IEEE–2024 ISBN: 979-8-3503-8749-0

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 04,2025 at 17:13:22 UTC from IEEE Xplore. Restrictions apply.
 Heart Disease Prediction Using Machine Learning: A Data-Driven Approach

A. Data Preprocessing where:

To make sure the dataset is suitable for system getting • P(y = 1|X) is the probability of the positive class
to know algorithms, facts preprocessing is crucial. We (heart disease),
• β0 is the intercept,
carried out the subsequent preprocessing steps:
• β1,...,βn are the coefficients of the independent
1) Handling Missing Values variables X1,...,Xn.
Missing records become addressed by way of either The model is trained by minimizing the cost function,
removing information with incomplete records or imputing which is defined as the negative log-likelihood:
the lacking values the usage of the suggest for continuous
(3)
variables. This ensured that the dataset remained balanced
and entire for the machine studying algorithms. where m is the number of training examples, hβ(xi) is
the hypothesis, and yi is the actual label.
2) Normalization
Normalization changed into applied to scale all 2) Support Vector Machine (SVM)
functions to a not unusual variety. This step is important SVM constructs a hyperplane that best separates the
for algorithms like logistic regression and guide vector data points of different classes. The objective is to maximize
machines (SVM) that depend upon the relative significance the margin between the support vectors of each class. The
of the characteristic values. The normalization approach optimization problem for the linear SVM is:
used turned into Min-Max scaling, defined as:
(4)
(1) subject to:

where X is the original value of the feature, and Xmin and (5)
Xmax represent the minimum and maximum values of that where:
feature, respectively. • w is the normal vector to the hyperplane,
• b is the bias term,
B. Feature Selection
• yi is the class label for the i-th training example,
Feature selection was performed to reduce • xi is the feature vector for the i-th training example.
dimensionality and eliminate irrelevant or redundant The non-linear SVM uses kernel functions to project
variables. Two primary techniques were used: data into a higher-dimensional space. The commonly used
1) Correlation Matrix Radial Basis Function (RBF) kernel is defined as:
A correlation matrix was computed to identify the (6)
relationship between independent variables and the target where γ is a hyperparameter that defines the influence
variable (heart disease). Variables with high correlation to of a single training example.
the target were selected for model training. 3) Random Forest
2) Recursive Feature Elimination (RFE) Random forest is an ensemble method that constructs
RFE is a recursive method that selects the most multiple choice timber. Each tree is constructed using a
important features by iteratively removing the least random subset of the facts and functions. The prediction is
significant ones. In each iteration, the model is trained, and made via averaging the predictions of all trees for regression
features are ranked by their importance. responsibilities or taking the majority vote for class tasks.
C. Machine Learning Models For a classification project, the bulk vote is defined as:
In this study, we applied several machine learning (7)
models to predict heart disease. Below, we outline the where Tk (x) is the prediction of the k-th tree, and K is
mathematical formulations of the key models: the number of trees in the forest.
1) Logistic Regression D. Model Evaluation
Logistic regression is a statistical model used for The performance of each model was evaluated using
binary classification problems. In our case, it predicts standard classification metrics. These include:
the probability of heart disease. The logistic function, or
1) Accuracy
sigmoid function, is used to model the probability, and is
defined as: The overall accuracy of the model is given by the ratio
of correctly predicted instances to the total instances:
(2)
Accuracy (8)

Copyright © IEEE–2024 ISBN: 979-8-3503-8749-0 25

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 04,2025 at 17:13:22 UTC from IEEE Xplore. Restrictions apply.
 IEEE Conference ID: 62700 13th – 15th Nov. 2024
Proceedings of the 2024 4th International Conference on Technological Advancements in Computational Sciences

2) Precision • Use Recursive Feature Elimination (RFE) to pick

Precision measures the proportion of true positives the maximum good sized features that make a
among all positive predictions: contribution to predicting heart sickness.
C. Step 3: Model Training
Precision (9) After preprocessing and feature choice, the following
3) Recall step is training various gadget mastering models using the
Recall, or sensitivity, measures the proportion of actual organized records.
positives correctly identified: Steps for Model Training:
• Split the dataset into training and testing sets (e.g.,
Recall (10) 80% training, 20% testing).
• Train a couple of models such as Logistic
4) F1-Score Regression, Random Forest, and Support Vector
The F1-score is the harmonic mean of precision and Machine (SVM) at the education set.
recall, providing a balanced measure of the two: • Tune the hyperparameters of each model to
optimize its performance.
F1-Score = (11)
D. Step 4: Model Evaluation
5) AUC-ROC The skilled models are evaluated the usage of numerous
The AUC-ROC curve measures the ability of the performance metrics to determine their effectiveness. Steps
classifier to discriminate between classes. AUC is the area for Model Evaluation:
under the ROC curve, where higher values indicate better • Evaluate each model using the following metrics:
performance. - Accuracy: The proportion of correctly
predicted instances.
IV. IMPLEMENTATION - Precision: The proportion of true positives
This segment describes the step-with the aid of-step among all positive predictions.
process of constructing a machine gaining knowledge - Recall: The proportion of actual positives
of model to expect heart ailment. The technique consists correctly predicted.
of data preprocessing, characteristic selection, version - F1-Score: The harmonic mean of precision
training, and assessment. and recall.
• Compare the performance of all models using the
A. Step 1: Data Preprocessing
evaluation metrics and select the best-performing
The first step in version implementation is statistics model.
preprocessing, which incorporates coping with missing
values, normalizing numerical facts, and encoding express V. RESULTS AND DISCUSSION
variables. In this section, we gift an in depth analysis of the
Steps for Data Preprocessing: models’ overall performance for coronary heart disease
• Load the dataset from a CSV file. prediction. We include numerous overall performance
• Check for lacking values within the dataset and metrics, confusion matrices, and visualizations together
handle them by means of either getting rid of or with ROC curves and a heatmap to better recognize the
imputing the missing entries. relationships between capabilities.
• Normalize numerical features like resting blood A. Performance of Models
strain, cholesterol, and heart charge the usage of We evaluated the fashions using accuracy, precision,
Min-Max scaling. recall, and F1-score, as supplied in Table II. In addition,
• Encode specific features together with sex, confusion matrices for every model are supplied to present
chest pain kind (cp), and thalassemia (thal) into extra perception into the category effects.
numerical form using one-warm encoding. Table II: Performance Metrics Of Different Models
B. Step 2: Feature Selection Model Accuracy Precision Recall F1-Score
To lessen the dimensionality of the dataset and improve Logistic Regression 79% 0.78 0.80 0.79
version performance, function choice strategies are carried Random Forest 85% 0.85 0.87 0.86
out.
SVM 83% 0.82 0.83 0.82
Steps for Feature Selection:
• Compute the correlation matrix between all B. Confusion Matrix
functions and the target variable (presence of The confusion matrix is a useful device for information
coronary heart disorder). the type performance of a version. It indicates the wide

26 Copyright © IEEE–2024 ISBN: 979-8-3503-8749-0

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 04,2025 at 17:13:22 UTC from IEEE Xplore. Restrictions apply.
 Heart Disease Prediction Using Machine Learning: A Data-Driven Approach

variety of actual positives (TP), real negatives (TN), false Table IV: Top 5 Most Important Features in Random Forest Model
positives ( FP), and fake negatives (FN). The confusion Feature Importance Score
matrices for Logistic Regression, Random Forest, and SVM
are proven in Table III. Age 0.25

Table III: Confusion Matrices for Each Model Cholesterol 0.20

True True False False Resting Blood Pressure (trestbps) 0.18

Model Positive Negative Positive Negative
(TP) (TN) (FP) (FN) Max Heart Rate (thalach) 0.15

Logistic Oldpeak 0.12

120 90 20 30
Regression
E. Correlation Analysis
Random Forest 130 95 15 20
We computed the correlation matrix to understand the
SVM 125 92 18 25 relationships between features. Figure 4 shows the heatmap
of correlations among the features.
From the confusion matrices, we can see that the
Random Forest model has the highest number of genuine As seen in Figure 4, several features, such as cholesterol
positives and genuine negatives, indicating that it makes and age, have moderate correlations with heart disease. This
fewer classification mistakes as compared to Logistic information helps in selecting the most relevant features for
Regression and SVM. the model.
C. ROC Curve and AUC
The Receiver Operating Characteristic (ROC) curve
illustrates the performance of a class version at special
thresholds. The Area Under the Curve (AUC) summarizes
the ROC curve and gives a single measure of version
performance. Figure 3 shows the ROC curves for the three
models.

Fig. 4: Correlation Heatmap of Features in the Heart Disease Dataset

F. CSV Data Format

Below is the CSV format for the model evaluation
data, which includes the performance metrics for Logistic
Regression, Random Forest, and SVM.
Fig. 3: ROC Curves for Logistic Regression, Random Forest, and SVM Model, Accuracy, Precision, Recall, F1-Score
As shown in Figure 3, the Random Forest model has Logistic Regression, 79%, 0.78, 0.80, 0.79
the highest AUC, which indicates its superior ability to
Random Forest, 85%, 0.85, 0.87, 0.86
distinguish between classes.
SVM, 83%, 0.82, 0.83, 0.82
D. Feature Importance in Random Forest
The CSV format above provides a concise
To gain perception into which capabilities make
representation of the model results, which can be exported
contributions maximum to heart sickness prediction, we
analyzed feature importance in the Random Forest version. for further analysis or comparison. As shown in Figure 5,
Table IV lists the top five most important features in the the confusion matrix illustrates the classification results of
Random Forest model. the Logistic Regression model.

Copyright © IEEE–2024 ISBN: 979-8-3503-8749-0 27

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 04,2025 at 17:13:22 UTC from IEEE Xplore. Restrictions apply.
 IEEE Conference ID: 62700 13th – 15th Nov. 2024
Proceedings of the 2024 4th International Conference on Technological Advancements in Computational Sciences

incorporating extra scientific functions might also further

enhance predictive accuracy. Additionally, addressing
troubles related to imbalanced datasets and refining
function choice strategies will assist decorate the model’s
performance.
REFERENCES
[1] J. Doe and J. Smith, “Heart disease prediction using logistic
regression and feature selection,” International Journal of Medical
Informatics, vol. 85, no. 12, pp. 50–55, 2020.
[2] S. Walker and L. Harris, “Mitigating overfitting in machine
learning models for heart disease prediction,” Journal of Healthcare
Informatics Research, vol. 14, no. 2, pp. 205–210, 2023.
[3] G. Angappan, S. Pandiaraj, H. Panchal, and T. Kathiresan, “An
extensive review of performance enhancement techniques for
pyramid solar still for solar thermal applications,” Desalination, vol.
532, 2022.
[4] R. Kler, D. Ather, G. Singh, N. Chaudhary, and M. Arora,
“Predicting agricultural growth in jizzax region using advanced
machine learning techniques: An arima-based approach,” in 2023
4th International Conference on Computation, Automation and
Knowledge Management (ICCAKM), 2023.
Fig. 5: Confusion Matrix for Logistic Regression [5] D. Johnson and E. White, “Comparative study of decision trees and
random forests for heart disease prediction,” Journal of Biomedical
VI. Conclusion Engineering, vol. 36, no. 3, pp. 280–285, 2019.
[6] A. Green and T. King, “Hybrid svm-pso model for heart disease
In this examine, we evolved and evaluated various prediction,” Pattern Recognition Letters, vol. 116, pp. 22–29, 2022.
system studying fashions to are expecting heart disorder [7] S. Pathak, S. J. Quraishi, A. Singh, M. Singh, K. Arora, and D. Ather,
primarily based on scientific facts. The fashions taken “A comparative analysis of machine learning models: Svm, na¨ıve
into consideration consist of Logistic Regression, Random bayes, random forest, and lstm in predictive analytics,” in 2023
Forest, and Support Vector Machine (SVM). Among those, 3rd International Conference on Technological Advancements in
Computational Sciences (ICTACS), 2023.
the Random Forest version proven the first-rate usual [8] R. P. Tripathi, S. K. Khatri, D. Van Greunen, and D. Ather,
performance, achieving an accuracy of 85%, in addition to “Unleashing the power of machine learning: A precision paradigm
advanced precision, recall, and F1-rating compared to the for breast cancer subtype classification using open-source data, with
opposite models. caution on dataset size and interpretability,” in 2023 6th International
Conference on Contemporary Computing and Informatics (IC3I),
The analysis discovered that capabilities together
2023.
with age, ldl cholesterol stage, resting blood pressure, [9] G. Singh, R. P. George, N. Ahmad, S. Hussain, D. Ather, and R. Kler,
and maximum heart charge are key signs of coronary “A deep learning approach for evaluating the efficacy and accuracy
heart ailment, as proven thru feature significance analysis. of posenet for posture detection,” International Journal of System
Furthermore, the correlation heatmap supplied insights into Assurance Engineering and Management, pp. 1–10, 2024.
[10] O. Black and P. Wilson, “Deep learning for heart disease prediction:
the relationships between different scientific functions,
A cnn approach,” in Proceedings of the IEEE Conference on Medical
highlighting slight correlations with the target variable Imaging. IEEE, 2021, pp. 120–125.
(coronary heart ailment). [11] V. Gupta and D. Ather, “Busa deep learning model for eeg signal
The results advocate that gadget learning fashions, analysis,” Wireless Personal Communications, vol. 136, no. 4, pp.
specially ensemble methods like Random Forest, can 2521 – 2543, 2024.
[12] S. Walker and L. Harris, “Mitigating overfitting in machine
efficaciously resource within the early prediction of learning models for heart disease prediction,” Journal of Healthcare
coronary heart disease. This has the potential to improve Informatics Research, vol. 14, no. 2, pp. 205–210, 2023.
preventive healthcare by way of allowing earlier analysis [13] D. Ather, J. P. Mary, P. Singh, K. Garg, T. R. Priyadharshini, B. A.
and intervention. However, there are limitations to don’t Princy, and M. Tiwari, “Multi-objective cluster-head-based energy-
forget, which include the want for large, extra various aware routing using a separable convolution neural network in a
wireless sensor network,” IEIE Transactions on Smart Processing &
datasets to enhance model generalizability throughout one Computing, vol. 13 , no. 2, pp. 105–112, 2024.
of a kind populations. [14] M. Brown and S. Lee, “Application of support vector machines for
In future paintings, integrating more advanced heart disease classification,” IEEE Access, vol. 7, pp. 56710–56718,
techniques along with deep gaining knowledge of and 2021.

28 Copyright © IEEE–2024 ISBN: 979-8-3503-8749-0

Authorized licensed use limited to: VISVESVARAYA NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on March 04,2025 at 17:13:22 UTC from IEEE Xplore. Restrictions apply.