Machine Learning Models Based on Grid-Search Optimization and Shapley Additive Explanations (SHAP) for Early Stroke Prediction

Ahstract- Timely detection of stroke is paramount for effective intervention and improved patient outcomes. This study introduces an innovative method for early stroke prediction utilizing machine learning (ML) models optimized through Grid-Search Optimization and elucidated using Shapley Additive Explanations (SHAP). The model not only identifies at-risk individuals but also enables timely preventative interventions. Twelve ML models were thoroughly compared and evaluated with the XGBoost model appearing as the superior performance across all metrics, achieving an accuracy of 94.29%, precision of 92.65%, recall of 96.19%, F1-score of 94.39%, and an impressive AUC ROC value of 99.82%. Furthermore, the SHAP analysis further enhances interpretability, highlighting elevated glucose as a major risk factor, along with age and BMI, aligning with established medical knowledge. This integrated ML methodology, combining performance-driven grid-search optimization and interpretability through SHAP, holds promise for assisting healthcare professionals in targeted stroke prevention strategies by accurately identifying high-risk individuals.