Using Machine Learning for Identification of Athlete Availability Predictors in a Multisport Elite Female Athlete...

Moore, SR, Cantú, EI, Brantner, CL, Britton, ME, DelBiondo, GM, Blue, MNM, Bruinvels, G, Hackney, AC, Register-Mihalik, JK, and Smith-Ryan, AE. Using machine learning for identification of athlete availability predictors in a multisport elite female athlete cohort. J Strength Cond Res XX(X): 000-000, 2026-This study used machine learning to identify influential predictors (training load, recovery, wellness, body composition, and force plate jumps [FP]) of athlete availability (AA) in female athletes. Fifty-two National Collegiate Athletic Association Division I female lacrosse (n = 28) and soccer (n = 24) athletes were tracked across a season for AA (% of unmodified practices and games). Potential predictors included measures of training load (average max speed, top speed, distance, high-intensity running [HIR], acute to chronic ratios for distance and HIR), sleep recovery (duration, efficiency, resting heart rate, heart rate variability, and temperature), wellness (stress, soreness, mood, fatigue, sleep quality, and hormonal profile), body composition changes (lean mass, percent body fat, fat-free mass index, and bone mineral density), and preseason FP testing (reactive strength modified, eccentric mean braking, jump height, and peak power). Influential predictors were identified using elastic net regression. Goodness of fit was described with root mean squared error (RMSE). Root mean squared errors for combined, lacrosse, and soccer models were 17.8, 8.9, and 17.8%, respectively. Influential predictors of training load, recovery, and wellness variables were selected in all models, with differing impact between teams. Single-team analyses demonstrated inconsistent predictors; most notable contrasts included no FP variables selected for lacrosse AA, and no body composition variables selected for soccer AA. Differences in RMSEs and predictors between teams suggest single-team models as a stronger approach to understanding key outcomes for AA. Comprehensive modeling to identify athletes at risk of low athlete availability before injury would be valuable for practitioners to refine testing batteries and hone meaningful predictors to prioritize data streams.