Cardiovascular Risk Prediction Using AI

Abstract

A silent epidemic, cardiovascular diseases (CVDs) continue to be the primary reason for loss of life around the globe. This emphasizes the critical need for early-stage risk prediction models for CVDs. This implementation aims to use an existing model for CVD risk prediction, which will be AI-driven and reliable. The main objective is to use machine learning techniques on medical datasets, it also focuses on transparency of the model along with user experience to ensure its implementation in the healthcare sector. The model utilized key clinical features that were selected using distinct feature selection methods, namely analysis of variance, mutual information, and chi-square. The feature groups selected were CRF-1, CRF-2, CRF-3, CRF-4, and CRF-5, respectively. The dataset was preprocessed with SMOTE (Synthetic Minority Oversampling Technique), which helps to resolve class imbalance. All machine learning algorithms were evaluated using Stratified K-Fold Cross Validation and Randomized Search Cross Validation. The final model's interpretability was enhanced by integrating Explainable AI (XAI) techniques, especially Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). Among the algorithms tested, the Multi-Layer Perceptron (MLP) Classifier with CRF-1 achieved an AUC of 0.98 with an accuracy of 96.72%. This makes the prediction more interpretable for the healthcare professionals and patients.