Optimizing Stroke Risk Prediction: A Machine Learning Pipeline Combining ROS-Balanced Ensembles and XAI

High accuracy alone isn't enough for clinical use: While the model achieved 99.09% accuracy, the researchers emphasized that interpretability through explainable AI (XAI) techniques was equally important for clinical applicability, demonstrating that doctors need to understand why a model makes predictions, not just trust its numbers.

Three key health factors drive stroke risk predictions: The study identified age, hypertension, and glucose levels as the most important variables for predicting stroke risk using LIME-based interpretability analysis, suggesting that AI systems can help clinicians focus on the most actionable health indicators.

Handling data imbalance is critical for fair predictions: The researchers used Random Over-Sampling (ROS) to address class imbalance in their dataset, highlighting that without this step, machine learning models can produce biased predictions that may miss high-risk patients or over-alarm low-risk ones.

Combining multiple AI models outperforms single approaches: The optimized ensemble combining Random Forest, ExtraTrees, and XGBoost performed better than individual models, suggesting that responsible AI implementation may require integrating diverse algorithmic approaches rather than relying on a single "black box" solution.

Explainable AI bridges the gap between accuracy and trust: By making the model's decision-making process transparent through XAI techniques, the framework enables healthcare providers to validate predictions against their clinical expertise and make informed, personalized treatment decisions rather than blindly following algorithmic recommendations.