Prediction of the Successfulness of Tennis Players with Machine Learning Methods

The aim of the study was to examine the predictability of the competitive performance of Slovene tennis players by using the most promising morphological measures and motor tests selected by automatic computer methods and by experienced tennis coaches by means of machine learning methods. The analysis included altogether 1,002 male and female tennis players who had undergone regular testing by the National Tennis Association and were positioned on the ranking list of the Slovene Tennis Association between the years 1993 and 2008. Selections of the most promising variables by means of the two automatic methods yielded similar results, whereas the selection performed on the basis of estimates made by coaches differed considerably. With regard to the analysis by means of classification methods, an accurate predictability of competitive performance for the age category younger than 16 years was observed, while the results of predictions for the age category older than 16 were poor. Among the regression methods, as opposed to the linear regression, which has yielded satisfactory results, regression trees served no useful purpose in practice. Automatic methods for identifying the most promising variables proved to be more successful than those of the coaches, which was most clearly noticeable with regard to the female tennis players and when linear regression was used.