A Systematic Comparative Study of Classical and Ensemble Learning Models for Scalable Cardiovascular Disease Prediction

Abstract

Cardiovascular disease (CVD) remains one of the leading causes of mortality worldwide, creating a strong need for accurate and scalable prediction systems capable of supporting early clinical diagnosis. Existing studies on cardiovascular prediction often focus on limited datasets or evaluate only a small number of machine learning models, resulting in restricted generalization and inconsistent comparative analysis. To address these limitations, this study presents a systematic comparative evaluation of classical machine learning algorithms, ensemble learning approaches, and hybrid deep learning techniques for cardiovascular disease prediction using a large-scale clinical dataset containing approximately 70,000 patient records. The study evaluates Logistic Regression, Decision Tree, Support Vector Machine, K-Nearest Neighbors, Gaussian Naïve Bayes, Multi-Layer Perceptron, Random Forest, Gradient Boosting, and XGBoost under identical preprocessing and experimental conditions. Data preprocessing includes feature scaling, encoding, data cleaning, feature importance analysis, and class balancing using SMOTE to improve prediction reliability and generalization. Model performance is evaluated using Accuracy, Precision, Recall, F1-Score, confusion matrix analysis, and comparative statistical ranking. Experimental results demonstrate that ensemble learning models consistently outperform traditional approaches, with XGBoost achieving the highest overall performance of 74.80% accuracy and strong recall performance for identifying cardiovascular risk cases. The findings confirm that ensemble learning methods provide improved predictive stability, reduced false negatives, and better clinical applicability for healthcare decision support systems. The study highlights the importance of standardized comparative evaluation frameworks for reliable cardiovascular disease prediction and demonstrates the effectiveness of ensemble-based learning for scalable medical analytics.