Abstract

Optimal athletic performance requires balancing intensive training with adequate recovery. Inertial Measurement Units (IMUs) provide objective data to monitor training loads, overcoming limitations of subjective assessments that risk overtraining. Using AI, this study analyzes IMU-derived multivariate time series data from 19 athletes, capturing triaxial acceleration, angular velocity, and magnetic orientation during standardized and fatigue-inducing sessions. A supervised machine learning model, comparing Random Forest, Gradient Boosting Machines, and LSTM networks, predicts fatigue and stamina with high accuracy. Real-time feedback and bias correction enhance model reliability, enabling personalized training adjustments that align with physiological thresholds. The model anticipates fatigue onset, reduces overtraining risks, and optimizes training periodization, improving performance. This AI-driven approach transforms sports analytics, supporting real-time monitoring and data-driven decision-making for tailored training and enhanced outcomes.