Cardiovascular Disease Prediction Using Machine Learning : A comprehensive Study

Shivank Kumar, Nisha Bisht, Naman Garg, Pranav Kumar, Imran Ansari
Computer Science Department, Greater Noida Institute of Technology, G.B Nagar,
Uttar Pradesh, India

Abstract - cardiovascular diseases (CVDs) long term health complications or death.

represent a significant global health Conventional diagnostic procedures,
challenge and need appropriate prediction through reliable, require complicated
models to make an early diagnosis and procedures, take time, and are very
treatment. The given study develops a resource-intensive.
machine learning based prediction model
Recent advances in machine learning
using data science techniques to identify
opened new frontiers in medical
the people who are exposed to CVDs risk.
diagnostics, providing powerful tools for
Different ML algorithms used for the
predicting risk to CVD based on many
prediction of a model are logistic
aspects, including clinical data, lifestyle
regression, random forests, gradient
habits and genetic predispositions. Some
boosting, and neural networks. The model
promising algorithms for processing big
was trained using clinical datasets
datasets and recognizing complex patterns
incorporating demographic information,
show that they can predict outcomes better
medical history, and lifestyle factors.
and faster than traditional methods of
Performance metrics include accuracy,
predicting CVD. With the help of the
precision, recall and F1-score for
health record of the patient, imaging data,
selecting the most appropriate algorithm.
and real-time biomarkers monitoring, the
The results indicate that when ML is
machine learning algorithms could
combined with appropriate data
uncover the subtle relation of the risk
processing techniques, there is a
factors involved with disease progression.
considerable improvement in prediction
That makes it an essential tool both for
accuracy, and this might find applications
clinicians and healthcare systems.
in preventive healthcare.
This research paper analyses the machine
learning technique in risk prediction for
I. Introduction cardiovascular disease. It analyses its
strength, challenges, and future
Cardiovascular diseases are still the
potentialities of this method. We examine
principal causes of death globally,
important algorithms such as decision
accounting for significant shares of global
trees, support vector machines, and neural
health burdens. Thereby, with continuously
networks on how effective they are at
growing risk factors such as sedentary
prediction for CVD.
lifestyles, unhealthy diets, smoking and
ageing populations, early diagnosis and Moreover, we inspect data quality, feature
forecast of CVDs can be crucial in further engineering, and model interpretability
decreasing its impact. Early diagnosis about the accuracy of predictions in
prevents the development of serious relation to improvement. Thus, with this
conditions like heart attacks, strokes, and new movement of the integration of
heart failure, which could lead to severe machine learning in clinical settings, its
potential will change the landscape by generalization, and reducing
which CVD will be diagnosed, managed, overfitting.
and prevented. This would find its way to
Another model that is popularly used is
improved health status worldwide.
Support Vector Machines, which have
demonstrated high accuracy in binary
classification tasks like predicting
II. Literature Review
whether a patient would suffer from a
Cardiovascular disease (CVD) prediction cardiovascular event.
has emerged as an important area of study
Moreover, Logistic Regression is still
in the health domain and with machine
one of the popular models for
learning (ML) offering innovative
predicting CVD risk.
solutions toward the improvement of early
diagnosis and preventive care. The
utilization of ML techniques in the
2. Deep Learning Models
prediction of CVD has garnered significant
attention due to the former’s ability to With more recent year’s focus being in
handle complex datasets, identify patterns, their capability to manage high
and predict outcomes with precision. This dimensions, deep learning refers to a
literature review aims at providing an form of machine learning using neural
overview of important studies that have networks with a minimum of two
implemented machine learning in the layers. Their applications have also
prediction of cardiovascular diseases: the been known in the interpretation of
methods, models, and datasets used medical images, from ECG and
therein, the contributions, and the future echocardiograms to the evaluation of
directions to be drawn. CVDs.
Machine Learning Models for CVD
Prediction
3. Challenges and Limitations
1. Supervised Learning Algorithms
This raises several challenges despite
Supervised learning algorithms are the optimism of machine learning-
perhaps the most popular in prediction based predictive models for CVDs.
of CVD. The algorithms learn from the One of the major limits is the quality
data with labels for classifying or of the data. Incomplete data or noisy
predicting an output for a given input data can readily hinder the
feature. A very famous one among performance of machine learning
them is decision tree and random models. Moreover, an absence of
forest. One of the early ones was by standardized datasets or protocols for
Zreik et al. In 2017 who utilized the data collection impedes the
decision tree for prediction of risk of generalization of results to different
CVD based on age, cholesterol levels, populations or settings. Another
and blood pressure. Random forests, a challenge is how to interpret complex
combination of multiple decision trees models especially deep learning
to improve predictive accuracy, have models, in that they are often
also been applied very successfully to conceived of as “black-box”. Deep
CVD prediction, providing better learning models may come with high
 predictive accuracy and do not always For outliers: use techniques like Z-
provide clearly understandable insights score, IQR (Interquartile Range) to
into the determinants of predictions, detect and handle.
hence important for clinical decision.
Data Splitting: Partition the data into a
training and testing set with 70-80%
III.Methodology for training and 20-30% for testing.

This approach includes several main steps 2. Feature Engineering

used to predict CVD via machine learning, Feature selection: Identify the most
which starts from data collection to pre- important features that significantly
processing, feature selection, model contribute to predicting cardiovascular
training, evaluation, and validation. The diseases. This can be done using:
remainder of this section will discuss in Correlation Analysis: check for
detail how the model building and correlations between features and the
validation are performed with evaluation target variable.
strategies and optimization towards
obtaining proper CVD prediction using Feature Importance: Use algorithms
models that are strong enough and like Random Forest to identify
generally applicable. important features.
Below is a step-by-step breakdown of the
process: - Feature Creation: Sometimes, new
features can be created, for example:
1. Data Collection and Pre-processing BMI (Body Mass Index) from height
Sources of Data: Gather data from and weight
publicly available datasets, such as the Cholesterol-to-HDL ratio
Framingham Heart Study, Cleveland Age adjusted risk factors
Heart Disease dataset, or clinical
databases. 3. Model Selection and Training
Logistic Regression: A simple and
Types of Data: The data usually interpretable model for binary
consists of structured records with classification.
features like:
 Demographics: Age, gender, Decision Trees/Random Forests:
ethnicity. Suitable for handling complex
 Medical History: Blood pressure, relationships and feature importance.
cholesterol levels, diabetes status.
 Lifestyle: Smoking habits, alcohol Support Vector Machines (SVM):
consumption, physical activity. Effective for high-dimensional spaces
 Clinical Measurements: Blood and non-linear decision boundaries.
pressure, ECG, heart rate.
K-Nearest Neighbours (KNN): A
Data Cleaning: Handle missing values, simple, instance-based learning
outliers, and errors in the dataset. algorithm.
For missing values: impute using
mean/median or use advanced
techniques.
 Neural Networks: If you have a large Predicted No Predicted
dataset, deep learning models can CVD CVD
capture complex patterns. Actual No True Negative False Positive
CVD (TN) (FP)
Model Training Actual CVD False Negative True Positive
Split the data into training (70%) and (FN) (TP)
testing (30%) sets to evaluate model Table 1: Confusion Matrix for
performance. Logistic Regression
Use cross-validation to tune model
hyperparameters and avoid overfitting. 2. Decision Tree
Decision Trees are understandable and
4. Model Evaluation interpretable models that recursively
 Accuracy: The proportion of divide the data into subsets based on
correct predictions. conditions of features. In, predicting heart
 Precision and Recall: Especially disease, decision trees can uncover
useful in imbalanced datasets. significant risk factors and offer insight
 F1-Score: A balanced measure of into process of decision making. They
precision and recall. suffer from overfitting, but this can be
 Confusion Matrix: To observe overcome using techniques such as
misclassifications and false pruning.
positives/negatives.
Performance Metrics:
 Accuracy- 85%
IV. Results and Conclusion  Precision- 0.75
 Recall- 0.70
1. Logistic Regression  F1-Score- 0.72
Logistic Regression is a base-level  AUC- 0.88
classification algorithm applied for binary
prediction. When applied to heart disease
prediction, it learns the probability of Predicted Predicted
No CVD CVD
occurrence from a linear combination of
Actual No 150 (TN) 30 (FP)
input variables. With its simplicity and
CVD
interpretability, logistic regression is a
good baseline model. It comes handy when Actual 40 (FN) 180 (TP)
investigating the association between CVD
independent variables and the risk of heart Table 2: Confusion Matrix for
disease. Decision Tree

Performance Metrics: 3. Support Vector Machine (SVM)

 Accuracy- 92% SVM is a robust classification technique
 Precision- 0.92 that operates by identifying a hyperplane
 Recall- 1.00 that maximally discriminates classes in the
 F1-Score- 0.96 feature space. SVM can deal with intricate
 AUC- 0.85 decision boundaries and is ideally suited
for high-dimensional datasets. In heart
disease prediction, SVM seeks to
determine an optimal boundary that
separate individuals at risk from those who Table 4: Confusion Matrix for
are not at risk. KNN

Performance Metrics: V. Future Scope

 Accuracy- 92%
The future scope of Cardiovascular
 Precision- 0.92
Disease Prediction Model using Machine
 Recall- 1.00
Learning is vast and include advancements
 F1-Score- 0.96
in technology, integration with healthcare
 AUC- 0.90
systems, and improvements in predictive
accuracy. Here are some key areas of
Predicted Predicted future development:
No CVD CVD
1. Enhanced accuracy with Deep Learning
Actual No 160 (TN) 20 (FP)
CVD Utilizing deep learning models like
CNNs and RNNs for ECG signal
Actual CVD 30 (FN) 150 (TP) analysis and time-series health data to
improve diagnostic precision.
Table 3: Confusion Matrix for
SVM 2. Integration with Wearable Devices &
IoT
4. K-Nearest Neighbours (KNN) Real-Time health monitoring using
KNN is a non-parametric classifier that smartwatches, fitness trackers, and
assigns data points to the majority class of IoT-enabled medical devices.
their k-nearest neighbours. In predicting
heart disease, KNN considers the Continuous data collection from
similarity of instances and hence is wearable sensors to predict
sensitive to local structures. Although cardiovascular risks dynamically.
KNN is computationally inexpensive, 3. Personalized and Adaptive Models
selecting an effective distance metric and
determining an optimal value for k are Developing models that adapt based on
important for its success. an individual’s genetic, lifestyle, and
environmental factors.
Performance Metrics:
4. Multi-Model Data Fusion
 Accuracy- 92%
 Precision- 0.92 Combining different data sources like
 Recall- 1.00 clinical reports, medical images (e.g.
 F1-Score- 0.96 echo diagrams), genetic data, and
 AUC- 0.86 patient history for more holistic
predictions.
Predicted Predicted 5. Real-Time Risk Prediction and Early
No CVD CVD Warning Systems
Actual No 140 (TN) 30 (FP)
CVD Deploying cloud-based AI systems that
can provide real-time alerts for patients
Actual 50 (FN) 180 (TP) at high risk of heart attacks or strokes.
CVD
 [6] Rahim A, Rasheed Y, Azam F, Anwar
MW, Rahim MA, Muzaffar AW (2021) An
integrated machine learning framework for
effective prediction of cardiovascular
VI. References diseases. IEEE Access 9:106575–106588.

[1] Vanisree K, Singaraju J (2011) https://doi.org/10.1109/

Decision support system for congenital ACCESS.2021.3098688
heart disease diagnosis based on signs and [7] Ashri SEA, El-Gayar MM, El-
symptoms using neural networks. Int J Daydamony EM (2021) HDPF: heart
Comput Appl 19(6):6–12. disease prediction framework based on
https://doi.org/10.5120/2368-3115 hybrid classifiers and genetic algorithm.
IEEE Access 9:146797–146809.
[2] Singh P, Singh S, Pandi-Jain GS (2018)
Effective heart disease prediction system https://doi.org/10.1109/
using data mining techniques. Int J ACCESS.2021.3122789
Nanomed [8] Khurana P, Sharma S Goyal A (2021)
https://doi.org/10.2147/ijn.s124998 heart disease diagnosis: performance
evaluation of supervised machine learning
[3] Li JP, Haq AU, Din SU, Khan J, Khan and feature selection techniques. In: 2021
A, Saboor A (2020) heart disease 8th International conference on signal
identification method using machine processing and integrated networks
learning classification in e-healthcare. (SPIN), Noida, India, pp 510–515.
IEEE Access 8:107562–107582.
https://doi.org/10.1109/
https://doi.org/10.1109/ SPIN52536.2021.9565963
ACCESS.2020.3001149
[9] Ishaq A et al (2021) Improving the
[4] Joo G, Song Y, Im H, Park J (2020) prediction of heart failure patients’
Clinical implication of machine learning in survival using SMOTE and effective data
predicting the occurrence of cardiovascular mining techniques. IEEE Access 9:39707–
disease using big data (nationwide cohort 39716.
data in Korea). IEEE Access 8:157643–
157653. https://doi.org/10.1109/
ACCESS.2021.3064084
https://doi.org/10.1109/
ACCESS.2020.3015757 [10] Nandy S, Adhikari M,
Balasubramanian V et al (2023) An
[5] Kavitha M, Gnaneswar G, Dinesh R, intelligent heart disease prediction system
Sai YR, Suraj RS (2021) heart disease based on swarm-artificial neural network.
prediction using hybrid machine learning Neural Comput Applic 35;14723-14737
model. In: 6th International conference on
inventive computation technologies https://doi.org/10.1007/s00521-021-06124-
(ICICT), Coimbatore, India, pp 1329– 1
1333.
https://doi.org/10.1109/
ICICT50816.2021.9358597