BACKGROUND: Predicting individual athletic performance presents unique challenges due to limited longitudinal observations per subject and the need for personalized rather than population-level inference. This study introduces a within-subject analytical framework combining log-linear growth curve models with Monte Carlo simulation for uncertainty quantification in swimming performance prediction. Unlike traditional time series approaches requiring extensive historical data, our method operates effectively with sparse longitudinal measurements (4-8 observations per athlete) by treating each individual as their own control; METHODS: We analyzed 417 kinematic observations (horizontal hip velocity) from 25 athletes (11 swimmers and 14 triathletes) across three swimming distances (200, 300 and 400 m), generating personalized forecasts with calibrated prediction intervals. The framework incorporates domain knowledge through the log-linear specification, which captures exponential skill acquisition dynamics established in motor-learning theory. RESULTS: Ninety five percent of athlete-distance combinations exhibited improving steps when assessed against individual baselines, with median growth rated of 4.93% per testing session. Monte Carlo simulation (10.000 iterations) produced well-calibrated 95% prediction intervals, with 92% of athletes showing greater than 50% probability of achieving personal bests in subsequent steps. The method successfully detected performance plateaus in 5% of cases and identified anomalous test results deviating from individual patterns. Breakpoint analysis revealed training response changes in 40% of steps. CONCLUSIONS: The individualized approach provides coaches with actionable, athlete-specific predictions and uncertainty estimated, avoiding the ecological fallacy inherent in between-group comparisons for individual sports performance optimization.
© Copyright 2026 XVth International Symposium on Biomechanics and Medicine in Swimming Proceedings. Alle Rechte vorbehalten.