Comparative Analysis of Computational Intelligence Techniques for Predicting Coronary Artery Heart Disease: A Study on Logistic Regression, Support Vector Machine, and K-Nearest Neighbor

Abstract

Aim: This study aimed to evaluate the performance of three machine learning models Logistic Regression (LR), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN) in predicting coronary artery heart disease using the Cleveland dataset. These algorithms were selected due to their interpretability, computational efficiency, and established utility as baseline classifiers in medical prediction tasks. Subject and Methods: The Cleveland dataset from the UCI repository was used, split into 90% training and 10% testing. Principal Component Analysis was applied for feature selection and stratified cross-validation helped address class imbalance. Models were evaluated based on accuracy, sensitivity, precision, F1-score, AUC-ROC, and confusion matrices. Hyperparameter tuning was performed using grid search with cross-validation. Results: LR outperformed other models, achieving the highest accuracy (90%), sensitivity (93.33%), and precision (87.50%). It also recorded the best AUC-ROC score (0.90) and effectively minimized both false positive and false negatives. SVM showed moderate performance (86.67% accuracy), while KNN was the least accurate (83.33%) with a higher false positive rate. Statistical significance testing (McNemar's test, p < 0.05) confirmed that LR's advantage over KNN was significant. Comparative analysis with recent studies indicated that the proposed approach performs competitively. Conclusion: LR proved to be the most effective model for heart disease prediction in this study. The use of PCA and stratified validation contributed to robust performance. These results support the role of interpretable ML models in clinical decision-support systems. Future work should include more diverse data sources, additional clinical variables, and explainable AI techniques to improve trust and deployment potential.