Explainable AI for Cardiovascular Health: A SHAP-Based Framework

In this paper an enhanced heart disease risk prediction system using binary classification with Explainable AI(XAI)based framework is proposed. With the initial data preprocessing and EDA, feature engineering is carried out to enhance the predictability of a given data set. The proposed model is trained using Extreme Gradient Boosting (XGBoost) and Random forest where XGBoost outperforms with an accuracy of 85 to 87 percentage and hence used for further development. Ensuring the model is actually making interpretable predictions, the (Shapley Additive Explanations) SHAP algorithm allows a more refined, visual form of feature contribution is implemented. A number of visualizations including interactive plots using Plotly and 3D scatter graphs helps to analyze the relationships between the features that helps to analyse contribution of each feature in SHAP algorithm to visualise using waterfall technique make to understand easily. In the final step, the model is deployed into an interactive, user-friendly Streamlit application built upon joblib and hence an actual tool may be used by people to assess their risk for heart disease.