Neural models as tools of sport prediction

One of most important problems of competitve sports is the proper selection of candidates for training. Its solution is above all related with the determination of such vector of variables (set of athlete personality features), which will bring the greatest quantity of information on the level of sports development of a given athlete. Another aspect of this problem concerns the determination of informative value (diagnostic value and prognostic value) of measuring tools (tests), which may be good indicators of the investigated features. The initial decision on the choice of candidates for sports training should be made on the basis of essential analysis of many features, among them somatic traits, efficiency and laboratory tests. This work shows that the tool assisting the candidate selection process for sports training may be neural-genetic model and therefore several possible solutions of this subject were proposed. Their construction and evaluation were made on the basis of two-year experimental studies with groups of young girl handball players. The results of our studies confirm earlier findings of Wit and Barton (1996), that neural networks may have special usefulness in theory of sports training. However, as has been shown earlier, the use of neural-genetic models for sport prediction requires relatively great number of observations and small number of variables.