Stability-XAI: Analisis Stabilitas Fitur dan Interpretabilitas Model Machine Learning pada Prediksi Penyakit Jantung

Ubaidillah Ubaidillah
Muhammad Taufik Roseno (Universitas Sumatera Selatan)
Serly Oktarina (Universitas Sumatera Selatan)

Algoritme Jurnal Mahasiswa Teknik Informatika
Vol 7 No 1 (2026): October 2026 || Algoritme Jurnal Mahasiswa Teknik Informatika

Heart disease requires early detection with high accuracy. This study proposes a Stability-XAI approach integrating feature stability analysis and Explainable Artificial Intelligence (XAI) to enhance the transparency of heart disease prediction models. Utilizing four UCI Machine Learning Repository datasets (Cleveland, Hungarian, Switzerland, Statlog) totaling 990 samples, the workflow includes preprocessing, feature selection (ANOVA F-score, Mutual Information, RFE), Optuna hyperparameter optimization across Logistic Regression, Random Forest, and XGBoost models, and SHAP interpretation. Results demonstrate that Optimized Random Forest achieves peak performance with 92.63% accuracy and 96.80% ROC-AUC. Features cp, ca, and exang exhibited the highest consistency (stability score 1.000), while cp, ca, and thal contributed most significantly according to SHAP analysis. The Stability-XAI approach effectively produces a heart disease prediction model that is high-performing, transparent, and interpretable.