Optimized Heart Disease Classification using Neural Networks and SMOTEENN Balanced Training

Heart disease remains one of the leading causes of mortality worldwide, underscoring the urgent need for accurate and early prediction systems. Traditional machine learning approaches often struggle with clinical datasets due to class imbalance—where instances of disease (positive cases) are significantly fewer than healthy controls. This study addresses this challenge by employing a hybrid resampling technique, SMOTEENN, which combines Synthetic Minority Oversampling Technique (SMOTE) with Edited Nearest Neighbors (ENN) to both balance the dataset and reduce noise. The proposed framework has been compared multiple models including Logistic Regression, Decision Tree, Voting Ensemble, and a Neural Network (MLPClassifier). These were trained on both the original imbalanced dataset and the SMOTEENN-balanced dataset. Experimental results showed that while models trained on the original dataset appeared accurate, they severely underperformed in detecting actual disease cases. In contrast, models trained on the resampled dataset achieved significantly improved sensitivity and generalizability. The Neural Network model trained on SMOTEENN data outperformed all others, achieving 93.3% overall accuracy, 96.4% recall, and an AUC of 0.963, making it highly suitable for clinical screening purposes.