XAI-reduct: accuracy preservation despite dimensionality reduction for heart disease classification using explainable...

Area of Science:

• Cardiology
• Artificial Intelligence
• Data Science

Background:

• Machine learning (ML) models are widely used for heart disease classification but often function as "black boxes", hindering interpretability.
• The "curse of dimensionality" poses a challenge, making resource-intensive classification necessary with comprehensive feature vectors (CFV).

Purpose of the Study:

• To reduce dimensionality in heart disease classification using explainable artificial intelligence (XAI).
• To maintain classification accuracy while improving model interpretability.
• To identify key features contributing to heart disease diagnosis.

Main Methods:

• Utilized four explainable ML models with SHapley Additive exPlanations (SHAP) for classification.
• Incorporated feature contributions (FC) and feature weights (FW) to generate a reduced dimensional feature subset (FS).
• Employed XGBoost classifier for its performance in explainable classification.

Main Results:

• XGBoost demonstrated superior heart disease classification accuracy with explanations, improving by 2% over existing methods.
• Explainable classification using the reduced feature subset (FS) outperformed most existing literature proposals.
• Accuracy was preserved with increased explainability using the XGBoost classifier.
• Identified the top four critical features for heart disease diagnosis, consistent across five explainable techniques applied to XGBoost.

Conclusions:

• Explainable AI, particularly using XGBoost with SHAP, effectively reduces dimensionality for heart disease classification without compromising accuracy.
• This approach enhances model interpretability and identifies crucial diagnostic features.
• The study represents a novel application of multiple explainable techniques to elucidate XGBoost for heart disease diagnosis.