An Explainable ConvNeXt-XGBoost Framework for Heart Disease Diagnosis Using SHAP and Counterfactual Explanations
Abstract
Heart disease remains one of the primary causes of mortality worldwide, creating a critical need for accurate and interpretable diagnostic systems that support timely clinical intervention. Although numerous machine learning and deep learning approaches have been developed for heart disease prediction, many existing models function as black-box systems and provide limited explanation regarding the factors influencing their decisions. This lack of transparency restricts their practical utilization in healthcare environments where interpretability is essential for clinical trust and decision-making. To address this challenge, this study proposes an explainable hybrid framework that integrates deep feature extraction, machine learning-based classification, and explainable artificial intelligence techniques for heart disease diagnosis. The proposed methodology employs a heart disease dataset consisting of 1,025 patient records and 14 clinical attributes. Initially, missing value handling, normalization, and standardization are performed to improve data quality and learning stability. Subsequently, ConvNeXt (Convolutional Next Generation Network) is utilized to extract deep informative feature representations from clinical data. The extracted features are then classified using XGBoost (Extreme Gradient Boosting) to predict heart disease occurrence. To enhance interpretability, SHAP (SHapley Additive exPlanations) is applied to quantify feature contributions, while DiCE (Diverse Counterfactual Explanations) generates actionable alternative scenarios for prediction analysis. Experimental results demonstrate an accuracy of 99.57%, precision of 99.62%, recall of 99.54%, and F1-score of 99.58%, confirming the effectiveness of the proposed framework for transparent and reliable heart disease diagnosis.