MODIFICATION OF THE MODULE BASED ON ARTIFICIAL NEURAL NETWORKS FOR PREDICTING RESULTS IN HIGHLEVEL SPORTS
- A.K. Krutikov
The development of information technologies and applied artificial intelligence makes it possible to improve the quality of
human work in various spheres of life. The sphere of physical culture and sports is no exception. Modern computer systems and
various machine learning methods are widely used for solving problems of statistical information processing, analysis, and planning
of both training and competitive activities.
The article analyzes the possibility of using various models of artificial neural networks for predicting sports events. The
structure of a modular forecasting system is described, where each module is implemented on one of the known models. The main
focus is on two models - based on a generalized regression neural network (GRNN) and based on a vector quantization neural
network (LVQ). The features of the formation of training choices and principles of work are indicated. As an example, we consider
the process of predicting the outcome of a Boxing match for the world champion title between Russian boxer Sergey Kovalev and
Mexican boxer Saul Alvarez. The results of the experiments are presented in tabular form. The analysis of these results allows us to
see the disadvantages of any of the models in predicting events in which, in addition to a monosyllabic answer (victory, defeat,
draw), it is necessary to obtain a number of numerical values.
To solve these problems, it is proposed to use cascading (pipelining) of system modules, each of which implements a particular
neural network model. A generalized structure of a two-level cascade system is presented. The use of this modification can be
considered successful, since even with a small expenditure on resources, this configuration showed the best forecasting results,
predicting the victory of Alvarez with almost 100% "confidence". However, the article notes that if the models used in cascading are
trained sequentially, this will lead to a significant increase in the time required to prepare the system for operation.
The prototype of the system was implemented on the basis of the MATLAB package at the V. A. Baykov Intelligent systems
research laboratory of the Vyatka state University. The system continues to be tested in more complex operating modes, with the
introduction of a large number of different types of source data and at different time intervals of forecasting. The use of the software
system by the coaching staff or personal trainers of athletes will help to increase the effectiveness of training, evaluate the level of
indicators of control tests to achieve the necessary competitive result.
- Index terms:
sports forecasting, artificial neural network, training sample, vector quantization, cascading, GRNN network, LVQ network, software system.