SubHealthAI: Predictive and Explainable AI for Early Detection of Subclinical Health Decline

This paper presents SubHealthAI, a predictive and explainable AI framework for early detection of subclinical health decline using wearable and lifestyle data. The system integrates Isolation Forest and GRU forecasting models with SHAP-based explainability and a Supabase audit pipeline for transparent, preventive AI insights. Evaluation on datasets such as WESAD and MIMIC-IV achieved Precision 0.81, Recall 0.76, RMSE 0.12, and MAE 0.09, demonstrating reliable early-detection potential for subclinical dysfunction. This version (v1.0, October 2025) corresponds to the initial public preprint aligned with the provisional patent “System and Method for Explainable AI Detection of Subclinical Physiological Dysfunction” filed with the USPTO. Future updates will extend the dataset (MIMIC-IV, UK Biobank), introduce LLM-based explainability, and enhance dynamic visualization through Apple Health–style UI components.