Advancing Heart Disease Prediction in App-Based Systems Using Optimized Machine Learning Model with Reduced Features and Explainable AI

Heart disease continues to be a leading factor in global mortality, emphasizing the importance of enhanced predictive healthcare systems. We developed an ML model to estimate heart disease risk based on personal medical information. After preprocessing and feature reduction, we applied nine ML models, including Random Forest (RF), Logistic Regression (LR), KNN, Decision Trees (DT), SVM, AdaBoost, Naive Bayes (NB), XGBoost, and Gradient Boosting (GB). Next, we removed the two least important features, fasting blood sugar (fbs) and resting electrocardiographic results (restecg). Following hyperparameter tuning, RF and XGB achieved 100% accuracy, but XGBoost was chosen as the optimal model because of its outstanding performance even prior to hyperparameter tuning. To enhance interpretability, SHAP and LIME were utilized. Finally, the XGBoost model was deployed on a web server and a mobile app, making it accessible for real-time heart disease predictions.