Optimizing Heart Disease Detection: An SVM-Based Methodology with Advanced Feature Selection

This study develops a clear framework to predict heart diseases using a Support Vector Machine (SVM) classifier. The methodology starts with overall data preparation which covers data cleaning which involves confirming the cleanliness of the data by clearing it of all smoothness and, handling of missing values which are solved appropriately, correcting data inconsistencies where such is the case and handling logical data errors if any. To address the class imbalance problem, the Synthetic Minority Over-sampling Technique (SMOTE) is used to create new instances from the minority class hence improving the balance of the data set. The high variability when creating the model is also handled through the usage of IQR which isolates the outliers and prevents excessive impact on the potential model. To enhance the relevance of the selected features and reduce multicollinearity Recursive Feature Elimination (RFE) and Correlation Coefficient analysis techniques are used. The accuracy, precision, recall and F1 scores for the SVM model applied to the SMOTE-balanced dataset with selected features are 98.24%, 89.43%, 93.64%, and 88.79% respectively. This also improves the accuracy of the heart disease model and, more importantly, makes the model clinically usable which involves retrieval drawbacks that are inherent to computational solutions as well as kernel selection.