Intelligent athlete evaluation: A real-time rating system based on TrueSkill-T and physical performance sensing

Abstract

Accurately evaluating the competitive level of football players is essential for optimizing training strategies and match performance. This study proposes an enhanced rating framework based on the improved TrueSkill-T algorithm to estimate players’ competitive levels over time. The model integrates match performance data with physical fitness metrics obtained via an embedded sensor network, forming a real-time, data-driven evaluation system. Key improvements include temporal smoothness constraints and robust handling of noisy and missing data. The system is implemented using Java, with modular architecture built around a microcontroller-based hardware setup. Experimental results on a dataset of over 18,000 match records demonstrate that the proposed method achieves faster convergence, lower rating errors (RMSE = 0.213), and greater resilience to data imperfections compared to baseline TrueSkill and Elo models. Furthermore, the system exhibits a significantly reduced response time and rating deviation, validating its suitability for real-world deployment in athlete performance monitoring and talent assessment.