Comparison of Various Computational Models for Accurate Stroke Prediction and Interpretation Using LIME and SHAP Kernels

Stroke is one dangerous illness that kills people. Stroke is so called because of the way it strikes people down. Across the globe, it is among the main causes of mortality and disability. It requires early identification and precise measures to be taken in order to prevent stroke and further damages to the brain and other parts of body. Our study investigates the use of machine learning techniques that includes Random Forest, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Free Classifier, Random Forest Classifier, Naive Bares and multiple Boosting Classifiers. The “Healthcare Dataset Stroke Data” is the dataset used, which is preprocessed and then trained upon by Ml. models. For Random Forest Classifier, the maximum accuracy observed was 94.91%, and upon Hyper parameter Tuning- there was rise in accuracy to 95.64%, Evaluation criteria like F1-score, precision, and recall provides additional evidence for the effectiveness of these models: Further in order to explain how each feature contributes differently to each prediction we have made use of two popular techniques for XAT, LIME and SHAP. LIVE and SHIP provides local, interpretable explanations and global explanations respectively.