Limitations of linear load-velocity modeling for bench press performance in youth elite athletes.

Introduction: Load-velocity profiling is widely used to characterize strength and estimate maximal load in resistance exercises. Most applications rely on linear regression models, assuming a linear relationship between load and velocity. This assumption has rarely been examined in youth elite athletes, who show pronounced inter-individual variability in neuromuscular coordination and technique. The primary aim was to examine whether linear load-velocity models describe bench press performance in youth elite athletes and whether a neural network approach better captures individual load-velocity characteristics. Methods: Fifty-three youth elite athletes completed one-repetition maximum testing and a standardized load-velocity protocol. Linear regression models were compared with neural network models for systematic bias, agreement with measured one-repetition maximum, and estimation error. Results: Linear models underestimated maximal strength and showed limited agreement with measured values, indicating structural limits in representing load-velocity behavior. Neural network models reduced bias and estimation error and indicated population-specific nonlinearity of the load-velocity relationship. The best-performing neural network showed high agreement with measured values with low absolute and relative errors. Discussion: These findings indicate that linear load-velocity assumptions may be insufficient for strength assessment in youth elite athletes and highlight the relevance of nonlinear behavior in this population.