An entropy weighted hybrid feature selection approach for heart disease prediction using machine learning

Heart disease still holds that top spot on the list of leading causes of death worldwide. Being able to predict heart disease at early stages really would help clinical practices work better and, in the end, it improves patient outcomes too. In this paper, we put forward an Entropy Weighted Hybrid Feature Selector (EWHFS), it is a kind of multi criteria feature selection framework, built with math backing, made for troublesome noisy and high dimensional biomedical datasets. The method blends Information Gain together with Fisher Score and also an Anti-Correlation Index, then it wraps everything into one entropy normalized optimization model. Overall, the model is set up to maximize feature relevance and class separability, but also reduce redundancy at the same time, not just one thing. We tested the approach on the UCI Heart Disease dataset, and we used stratified 10 fold cross validation, as usual. The experimental results show that EWHFS picked an optimal subset of 8 very discriminative features from 13 original clinical attributes. This cuts the dimensionality by 38.5% while still keeping the diagnostic information. So, the proposed approach reached about 94% accuracy, 93% precision, and 94% recall, and the F1 score came out as 93.5% while AUC was 0.95. It basically did better than the usual feature selection strategies, like Information Gain, Chi Square, Mutual Information, and even Recursive Feature Elimination. When we compare it with the strongest baseline, EWHFS bumped the classification accuracy up roughly 3–6%, yet it still kept pretty steady behavior across the validation folds. Then for the statistical validation, paired t tests plus Wilcoxon signed rank tests were used, and they verified that the gains were significant, with p < 0.05. These results demonstrate that EWHFS provides an effective, interpretable, and computationally efficient solution for heart disease prediction and clinical decision support systems.