Explainable AI for Cardiovascular Disease Prediction Using Electronic Health Records

Cardiovascular diseases (CVDs) are the major diseases in terms of morbidity and mortality at an international level, and there is a great necessity to prevent them in their early stages. The AI techniques have proved to be utilized to predict the cardiovascular risk factors, nevertheless, they are not frequently applied due to the inability to be easily comprehended by a specialist. This paper presents an Explainable Artificial Intelligence (XAI) system of predicting cardiovascular disease relying on Electronic Health Records (EHRs) to enhance accuracy and transparency. The datasets of publicly available EHRs are analyzed, such as demographic, clinical, and lifestyle variables, such as age, sex, blood pressure, cholesterol, glucose, and smoking. Training, the optimization and testing of some of the supervised learning models, including the Logistic Regression, Random Forest and XGBoost, was conducted on the basis of the measure of accuracy, F1-score and ROC-AUC. To resolve the interpretability issue, the framework integrates SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) to provide explanations to global and local predictions. As depicted in the experimental results, the most accurate was the XGBoost and the SHAP visualization revealed that systolic blood pressure, cholesterol level, and age had the greatest predictors of CVD. The findings have confirmed that the proposed XAI framework is not just related to an increase in the predictive reliability but it provides a clinician with clear understanding of model decisions. Such explainable models may play an important role in the risk stratification process, individualised treatment, clinical decision support and ultimately to precision medicine and trust in AI-based healthcare systems.