Autoregressive Time-Series Analysis of Biomechanical Anomalies: Insights into Exercise-Induced Fatigue in Healthy Participants.

Most fatigue-detection approaches for biomechanics rely on computationally intensive black-box models or apply thresholding without correcting for serial dependence, risking inflated false alarms when signals are autocorrelated. This study proposes an autoregressive integrated moving average with statistical process control (ARIMA-SPC) framework that models temporal dependence and identifies fatigue-related biomechanical anomalies from time-normalized ground reaction force and tibial inertial measurement units waveforms. In 32 participants performing repeated 90° lateral cutting maneuver trials before and after a 5-km variable-speed fatigue protocol. This is evident from their autocorrelation function plots, which oscillate and decay exponentially, forming a characteristic "tail-off" pattern. The most compelling finding is that the lag value of the coronal plane with ground reaction force is markedly greater than those of the other two planes (p<0.05). Relative to the long short-term memory model, the ARIMA-SPC model was easier to interpret and delivered stronger anomaly-detection performance, particularly in sensitivity (recall 0.88 vs 0.80; precision 0.92 vs 0.85; F1 score 0.90 vs 0.82). The ARIMA-SPC approach also demonstrated substantially lower inference time (5 s vs 15 s), indicating low computational burden compatible with real-time monitoring constraints. The findings of this study have practical applications in the development of real-time monitoring systems to detect exercise-induced fatigue. And prewhitened residual monitoring provides an interpretable and computationally efficient route for fatigue-related biomechanical anomaly detection.