COMPARATIVE ANALYSIS OF PERFORMANCE EVALUATION FOR STROKE RISK PREDICTION BASED ON CLINICAL DATA

Abstract

Stroke is one of the leading causes of death and disability worldwide, requiring an accurate machine learning-based risk prediction approach to support early detection. This study aims to conduct a comparative evaluation of three supervised learning algorithms, namely Naïve Bayes, Random Forest, and SVM, in predicting stroke risk. The clinical dataset used consisted of 5,110 patients. Model evaluation was performed using the Stratified 5-Fold Cross Validation method, a cross-validation technique that divides the data into five subsets while maintaining the class proportions in each fold. Each subset is alternately used as test data, while the other subset is used as training data, so that all data can be used as training data and test data. Model performance was measured using accuracy, confusion matrix, and AUC-ROC metrics to assess classification performance. The results showed that Random Forest achieved the best performance with an accuracy of 95%, followed by Naïve Bayes at 86% and SVM at 75%. Based on the AUC-ROC evaluation, Random Forest also showed the most optimal performance with a value of 0.80, indicating excellent classification ability. Random Forest is the most effective algorithm in predicting stroke risk in the dataset used, so it has the potential to be the best method used to support the early stroke detection system.