Comparative Evaluation of Machine Learning Models for Stroke Prediction in Clinical Settings

Abstract

Stroke is a leading global cause of death and disability, emphasizing the need for early and accurate prediction to improve patient outcomes. Traditional diagnostic methods often face limitations in early detection, highlighting the demand for advanced predictive tools. This study addresses this challenge by developing a Machine Learning (ML) based system for stroke prediction using clinical data. Five ML algorithms, namely, Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Logistic Regression (LR), and K-Nearest Neighbors (KNN), were evaluated on a dataset of 5,110 records with 10 key attributes, including demographic, physiological, and lifestyle factors. Rigorous preprocessing, including handling missing values, categorical encoding, and feature scaling, was applied to ensure data quality. Experimental results demonstrated that ensemble methods outperformed other classifiers, with RF achieving the highest accuracy (97.85%), precision (97.9%), recall (97.85%), and F1-score (97.59%). DT also exhibited strong performance (96.7% accuracy), while linear models (LR, SVM) and KNN showed limitations in handling dataset complexities. The study underscores the superiority of tree-based ensemble methods for stroke prediction, offering a reliable, interpretable framework for clinical decision-making. These findings highlight the potential of ML in enhancing early stroke detection and supporting timely interventions.