AUTOMATED STROKE PREDICTION USING MACHINE LEARNING AN EXPLAINABLE AND EXPLORATORY STUDY WITH A WEB APPLICATION FOR EARLY INTERVENTION

Abstract

Stroke poses a significant worldwide threat with severe health and economic implications, resulting from disruptions in blood flow to the brain and causing neurological impairment. As the aging population increases, the number of people at risk for stroke grows, emphasizing the urgent need for effective prediction systems. This project is addressing the challenge of stroke by developing automated prediction algorithms. These algorithms aim to enable early intervention, potentially saving lives by predicting strokes accurately. The precision and effectiveness of such systems become increasingly crucial in managing the rising population at risk. The project involves a comprehensive examination, comparing the effectiveness of a proposed machine learning technique with six well-known classifiers. Metrics related to both generalization capability and prediction accuracy were scrutinized to evaluate the performance of the developed algorithm in stroke prediction. To provide transparency into the black-box nature of machine learning models, the study employs explainable techniques, specifically SHAP (Shapley Additive Explanations). This method is well-established in the medical industry, offering insights into model decision-making processes. The experimental results indicate that more intricate models outperformed simpler ones, higher accuracy. The proposed framework, incorporating both global explainable methodology, aims to standardize complex models. This standardization can enhance stroke care and treatment by providing valuable insights into the decision-making process of the algorithms. It includes ensemble methods such as Categorical Boosting and Stacking Classifier were applied, leveraging the combined predictions of multiple individual models to enhance overall prediction accuracy. Notably, the Stacking Classifier demonstrated exceptional performance, achieving an impressive 99% accuracy.