HXAI-ML: A hybrid explainable artificial intelligence based machine learning model for cardiovascular heart disease detection

Abstract

Cardiovascular diseases (CVDs) are a leading cause of morbidity and mortality globally. Early diagnosis and accurate prediction are critical for effective prevention and treatment. However, traditional machine learning (ML) models for CVD prediction face challenges such as data imbalance, lack of interpretability, and limited generalization across datasets, which restrict their practical application in healthcare. This study introduces a hybrid explainable artificial intelligence-based ML (HXAI-ML) model to address these limitations. The proposed framework combines advanced data balancing techniques, including Random Oversampling (RO), Synthetic Minority Oversampling Technique (SM), RO+Tomek Link (TL), SM+TL, RO+Instant Hardness Threshold (IHT), and SM+IHT, with XAI tools like SHapley Additive exPlanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Permutation Importance Analysis (PIA). This integration enhances prediction accuracy while ensuring model transparency, allowing healthcare professionals to interpret decision-making processes effectively. The HXAI-ML model was validated on two benchmark datasets, Cleveland and Framingham, achieving superior performance. Combining TL and RO with the Extra Trees Classifier (ETC) yielded the highest results, with 98.78% accuracy and 98.82% precision on the Cleveland dataset, and 99.24% accuracy and precision on the Framingham dataset. These findings demonstrate the model's ability to outperform traditional approaches in both accuracy and interpretability. This research provides a scalable, clinically interpretable solution for early CVD detection, capable of improving decision-making and enabling personalized treatment strategies. The hybrid model represents a significant step toward integrating advanced ML into practical healthcare applications, improving patient outcomes. • HXAI-ML model improves CVD prediction with 98.78% accuracy on Cleveland dataset. • Combines data balancing and explainable AI for accurate and interpretable results. • Achieves 99.24% accuracy and precision on the Framingham dataset with hybrid techniques. • Ensures clinical transparency using SHAP, LIME, and PIA for decision interpretability.

Ähnliche Arbeiten