Personalized Neural Modeling for Daily Injury Risk Assessment via Wearable Health Data

Despite advances in wearable technology, existing models for athletic injury prediction often lack personalization and proper temporal alignment, limiting their effectiveness. In this work, we introduce the Personalized Athlete Injury Risk Model (PAIR), a neural network designed to predict daily self-reported injury risk scores using wearable sensor data. Evaluations on 36 collegiate athletes across 3,000+ daily observations show that PAIR achieves an R-squared value of 0.506, outperforming a non-personalized baseline (0.302) and highlighting the benefits of our approach. Our key contributions include: (1) developing a personalized neural network model that captures athlete-specific patterns through individualized embeddings and advanced signal processing, and (2) demonstrating that personalization and robust temporal alignment significantly improve prediction performance and utility.