Machine Learning-Based Stroke Prediction with Efficient Feature Importance Analysis

Abstract

A stroke represents a critical neurological and vascular emergency that occurs when blood supply to the brain becomes interrupted or reduced, triggering a complex cascade of events that lead to ischemic injury, oxygen deprivation, and ultimately, neuronal death. Advanced artificial intelligence and machine learning algorithms have revolutionized stroke prediction and diagnosis by analysing complex medical imaging data, patient histories, and clinical parameters with remarkable accuracy, enabling healthcare providers to make faster and more precise diagnostic decisions while identifying high-risk patients before stroke occurrence through pattern recognition in large datasets. This research paper explores the application of machine learning in stroke prediction, focusing on the identification of key risk factors. By utilizing a comprehensive dataset and employing a range of machine learning models, including logistic regression, decision trees, random forests, support vector machines, K-nearest neighbours, and gradient boosting, the study aims to uncover significant predictors of stroke. The primary objective is not to outperform existing models, but to gain a deeper understanding of feature importance in stroke risk assessment. Through a multifaceted approach to feature importance analysis, including built-in metrics for tree-based models, coefficient analysis for linear models, and permutation importance for other algorithms, the research identifies the most influential factors in stroke prediction. The findings of this study can contribute to improved stroke prevention and early detection by providing clinicians with interpretable, AI-assisted insights for informed decision-making.