Interpretation and Analysis of Machine Learning Models for Brain Stroke Prediction

Abstract

Brain stroke is a significant cause of death nowadays. As per WHO, 11% of the population dies yearly from it. However, early measures can save many lives. Machine Learning (ML) is used as a tool for early predictions in a human through their symptoms, lifestyle, and from medical history. With the advancement in machine learning, features responsible for brain stroke can be identified and ranked as per their effect. Such features for brain stroke are hypertension, smoking status, heart disease, body mass index, and sugar level. In this paper, various ML classifiers such as Neural Network (NN), Support Vector Machine (SVM), Random Forest (RMF), Decision Tree (DST), and Gradient Boost (GBST) are used to classify patients with brain stroke. The models are then compared for the best results. Lastly, Local Interpretable Model-agnostic Explanation (LIME) and SHAP (SHapley Additive exPlanations) are used for explanation to find the reason behind the decision taken by the best ML model. The results show that RMF (GBST after that) achieves the highest prediction accuracy.