Enhanced Stroke Prediction through Recursive Feature Elimination and Cross-Validation in Machine Learning

Abstract

Stroke is the primary cause of morbidity and mortality on a global scale, and early identification of people at high risk for stroke is critical for its effective prevention and treatment. On consuming clinical and demographic data, machine learning algorithms have shown the ability to predict stroke risk. This study developed and evaluated four stroke prediction models based on machine learning: Random Forest, SVM Classifiers, Neural Networks, and Logistic Regression. Various feature selection techniques were implemented, including correlation-based, mutual information-based, wrapper-based, and principal component analysis. Results showed that all four machine learning models achieved high accuracy rates, ranging from 85% to 91%. The Neural Network model had the highest accuracy at 91%. Random Forest had the highest sensitivity and specificity at 85% and 91%, respectively. The study demonstrates that machine learning models may be valuable tools for identifying individuals at high risk of stroke and that feature selection techniques can significantly impact model performance. The findings depict the development of more accurate and effective stroke prediction tools, ultimately improving stroke prevention efforts and reducing the burden of stroke on individuals and healthcare systems.