Improved Adaptive Kernel-Based SVM Ensemble for Enhanced Heart Disease Diagnosis: A Feature Optimization Approach

Since cardiovascular disease continues to be the world’s leading cause of death, precise, data-driven diagnostic systems must be developed. By combining multi-phase feature selection, adaptive model training, and sophisticated preprocessing, this study suggests an intelligent classification framework for the diagnosis of heart disease. Using a novel hybrid kernel function for dynamic feature space adaptation, the Improved Adaptive Support Vector Machine (IASVM) is an improved version of the Random Forest (RF), Neural Network (NN), and Support Vector Machine (SVM) classifiers. The dataset is first normalized for consistency and cleaned using outlier detection (Isolation Forest). A comprehensive method that combines correlation analysis, ANOVA F-test, Random Forest Importance (RFI), and SHAP value analysis is used to select features. These complementary methods reduce dimensional complexity while identifying a small, highly impactful subset of features that maintain diagnostic significance. To assess how feature reduction affects performance, all four models are trained on both the full feature set and the optimized subset for classification. The IASVM uses an adaptive kernel function that dynamically scales according to the geometric properties of the dataset, integrating the Manhattan and Euclidean distances. Because of this, it can more effectively differentiate between overlapping classes in nonlinear spaces, especially in clinical settings where feature distributions are unbalanced or skewed. To fine-tune classification boundaries, grid search and cross-validation are used in the IASVM’s hyperparameter tuning process. A real-world dataset of heart disease is used for experimental evaluations, and metrics like accuracy, precision, recall, F1-score, and AUC are used to compare the models. The findings highlight the IASVM’s capacity to generalise in high-dimensional and noisy domains by showing that, even though traditional models function well, it invariably performs better than the others, particularly when using the chosen feature set. Additionally, to improve clinical interpretability and usability, a graphical user interface (GUI) is created to display feature selection results, classification metrics, and accuracy comparisons. This study demonstrates how well feature optimisation and adaptive kernel-based classification work together, offering a scalable and understandable method for improving cardiovascular healthcare decision support.