A Comparative Study on Feature Selection between Computational and Medical Knowledge Driven Approaches for Heart Disease Prediction

Heart disease is a major cause of mortality globally, emphasizing the need for timely and accurate diagnostic methods. Machine learning algorithms, especially those utilizing artificial intelligence, offer promising solutions for heart disease prediction. However, medical datasets’ high dimensionality complicates identifying key risk factors for accurate classification. This study tackles these challenges by comparing various feature selection techniques, including Wrapper and Filter methods, and expert-driven approaches to identify the most relevant features from heart disease datasets. The goal is to reduce feature numbers while maintaining prediction accuracy. Machine learning classifiers like Support Vector Machine (SVM), Naïve Bayes (NB), Decision Tree (DT), K-Nearest Neighbor (KNN), Random Forest (RF), and Adaptive Boosting (AdaBoost) were applied to both full and reduced feature sets. Feature selection methods such as Sequential Forward and Backward Selection, Chi-Square, and Mutual Information were evaluated for their effectiveness. The models’ performance was measured using accuracy. The findings show significant improvements in classification accuracy and reduced training time with a reduced feature subset, validating the effectiveness of feature selection in heart disease prediction.