Evaluating Machine Learning and Deep Learning Techniques in Stroke Risk Prediction

Abstract

The current study explores the application of Machine Learning (ML) and Deep Learning (DL) techniques to predict stroke risk, addressing major challenges such as model accuracy, adaptability, feature importance, transparency, and external validation. A quantitative approach is used to evaluate various ML algorithms, and in particular, Random Forest has been highlighted because of its superior predictive accuracy, while stressing the adaptability of DL models across different demographic contexts. The study further explores the role of feature significance in enhancing context-specific predictions, the challenges of model explainability in clinical adoption, and the critical importance of external validation in ensuring generalizability. The results underline the transformative potential of ML and DL in advancing personalized healthcare strategies for stroke prediction while identifying existing gaps in transparency and validation practices. This synthesis lays a foundation for future research to standardize external validation protocols and improve model transparency for wider clinical adoption.