This research presents the selected statistical learning methods in predicting the results of 400 m hurdles in two different time intervals. The calculated models predict results in selected training period and in annual training cycle. In the study, detailed training programs of 21 Polish hurdlers were analyzed. Building of the predictive models was conducted by means of regression shrinkage and artificial neural networks. To evaluate calculated models the leave-one-out cross validation was used. The outcome of the studies shows that the best method in both analysed time intervals was LASSO regression. The prediction error for a training period was at the level of 0.67 s, whereas for the annual training cycle was at the level of 0.39 s. Additionally, for both time intervals the optimal set of predictors was calculated. In terms of training periods, the LASSO model eliminated 8 variables, whereas in terms of the annual training cycle 12 variables were eliminated.
© Copyright 2015 Sports Science Research and Technology Support: Second International Congress, icSPORTS 2014, Rome, Italy, October 24-26, 2014, Revised Selected Papers. Published by Springer. All rights reserved.