Sustainable and interpretable heart disease prediction: a clinical decision support approach for biomedical healthcare applications

Cardiovascular disorders cause approximately 18 million deaths annually worldwide, underscoring the urgent need for precise and rapid diagnosis. Conventional machine learning does not automate feature extraction and does not capture complex non-linear relationships in high-dimensional medical data during cardiodiagnostic predictions, not to mention current techniques also lack interpretability and transparency and thus are not useful for building clinically trusted AI-powered prediction tools for personalized biomedicine and healthcare. To fill this gap, we propose an interpretable Convolutional Neural Network (1D CNN) model for cardiodiagnostic predictions that integrates automated feature extraction and explains AI. Our approach uses a 1D CNN model composed of two convolutional layers (with 64 and 128 filters). The CNN will be trained on Cleveland Heart Disease Dataset (Kaggle) (303 instances, 14 attributes) and will undergo evaluation using accuracy, precision, recall , F1 score, and LIME-SHAP interpretability analyses. The results given attest to the remarkable results, where the model achieved 98.05% percent accuracy, 100% percent precision, 96.12% percent recall, 98.02% percent F1 score, 0.963 MCC, and 0.961 Kappa coefficient. Our results exceed several of the more modern techniques offered in the literature. LIME and SHAP analyses reveal how specific features (sex, number of major vessels, thalassemia status) drive model predictions, enhancing interpretability and aligning with clinical understanding of cardiac risk factors essential for precision medicine. This research demonstrates the potential for interpretable deep learning to transform cardiovascular diagnostics through enhanced clinical decision support systems and trustworthy AI implementation in precision medicine.