A Review on AI-Based Risk Stratification and Explainable Models for Wearable Neurological Health Monitoring Systems

Epilepsy, Parkinson's Disease, Alzheimer's Disease, and stroke are major contributors to disability or handicap on a global basis both clinically and economically; furthermore, while wearable sensor technology with artificial intelligence (AI) has the potential to monitor continuously and live outside the clinic, the use of black-box deep learning models continues to struggle due to the concerns of lack of transparency, lack of accountability, and inadequate regulatory compliance. This paper becomes a systematic review of AI-based risk stratification and explainable AI (XAI) models such as SHAP, LIME, Grad-CAM, attention-based visualizations, and layer-wise relevance propagation across the four neurological disease areas of epilepsy, Parkinson's, Alzheimer's/dementia, and stroke, categorizing the types of sensors and models used and the associated measured metrics of performance. Although wearable, AI-based devices are typically very accurate (e.g., ≥ 95% accuracy for detecting seizures), they are still faced with challenges in real-life implementation, such as regulatory compliance, energy efficiency, standardization of XAI evaluation systems, and generalizability to new data; future efforts will focus on multimodal sensor fusion, foundational models, and human-centered explainability within the clinical workflow.