Yu.Yu. Petrunin

There are new approaches to the study of sport in the late 20th century. New approaches the study of sport actively use methods of artificial neural networks. Application of neural networks in the sport is divided into three groups. The first group creates a causal model of athletic performance. The second group creates predictive models of the outcome of games. In these studies MLP, RBF, PNN, GRNN models of neural networks are used. Accuracy of prediction of the outcome of games with the help of neural networks in basketball and American football ranged from 74% to 81%. Accuracy of prediction of the outcome of games with the help of neural networks in European football is from 64% to 76%. Neural networks, according to the author, can also be used to detect the «match-fixing». The third groups of research use the models of self-learning neural networks (SOM, DyCON). The purpose of this research is to reduce the dimensionality of the empirical data, tactical pattern recognition in sport. The results of these studies allow simulating the tactics of teams in team games (e.g., in football). The use of self-organizing neural network models opens the hidden patterns in the evolution of world football. The article shows the growing influence of the theory of artificial neural networks in sports science in recent years. The author notes the expansion of the research themes of application of neural networks in sports science. The author also notes the diversity of models of neural networks used in sports science. The greatest efficiency, according to the author, reveals the use of the methods of neural networks with fuzzy logic models and techniques of genetic algorithm.