Predicting Cerebral Stroke: Comparison Between Anomaly Detection Algorithms and Classification Models

Cerebral stroke occurs due to hindered blood flow to various parts of the brain, often caused by ischemic blockages or haemorrhagic vascular rupture. These disruptions can lead to severe brain damage or death if not addressed promptly. Annually, over 15 million people worldwide suffer from cerebral stroke, with lifestyle factors such as hypertension and smoking playing a significant role. This study presents a comparative analysis of cerebral stroke prediction models utilizing classification algorithms—Random Forest, XGBoost, Gaussian Naïve Bayes, and KNN—as well as anomaly detection methods such as Local Outlier Factor, Isolation Forest, and Gaussian Probability Distribution. Among these, the Isolation Forest algorithm demonstrated superior performance, achieving a weighted TPR-TNR score of 0.7449, reflecting balanced anomaly detection capabilities in imbalanced datasets. The findings highlight the importance of tailored machine learning solutions for stroke risk prediction. Future work will explore hybrid approaches to enhance detection accuracy while addressing dataset challenges.