Stroke Risk Prediction based on SVM and XGBoost Algorithms with SMOTE-ENN

Abstract

Stroke is a neurological disorder that requires early risk identification. One of the challenges in stroke prediction using machine learning is class imbalance, which can cause models to favor the majority class. This study aims to compare the performance of Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost) for stroke prediction using the Synthetic Minority Over-sampling Technique–Edited Nearest Neighbours (SMOTE-ENN). The dataset used was the Stroke Prediction Dataset, consisting of 43,400 records and 12 attributes. The research procedure included data preprocessing, an 80:20 train-test split, application of SMOTE-ENN to the training data, SVM and XGBoost model training, and evaluation using accuracy, precision, recall, F1-score, confusion matrix, ROC-AUC, and computational time. The results show that SMOTE-ENN successfully increased the representation of the stroke class from 1.8% to a balanced distribution in the training data. The SMOTE-ENN-XGBoost model achieved an accuracy of 96.96%, precision of 96.83%, recall of 97.11%, and F1-score of 96.97%, with a training time of 4.68 seconds. In comparison, the SMOTE-ENN-SVM model achieved an accuracy of 90.13%, precision of 88.07%, recall of 92.83%, and F1-score of 90.39%, with a training time of 21.2 minutes. These results indicate that SMOTE-ENN-XGBoost provides better classification performance and computational efficiency than SMOTE-ENN-SVM for stroke prediction. The findings suggest that applying SMOTE-ENN with XGBoost can support stroke risk detection with high classification performance and more efficient computational time.