Methods for determining the five personality traits and character type of athletes using neural networks for sports selection

Abstract

Relevance. The problem of sports selection is constant and difficult to solve. This can be saddled if you determine their characteristics in terms of personality structure (Big Five) and then look for applicants with a similar configuration. However, it is well known that the reliability of written tests is not higher than 40%.

Purpose. We have written a software embodiment of a neural network in order to determine the top five personal qualities and type of character from a photo of the athletes' faces.

Methodology and organization of the research. A specially prepared dataset of more than 20,000 photos was used to train the network. During 20 eras of training, they received the lowest error rate and the highest proportion of correct answers -29.5%, while athletes are accurate only in 25% of cases.

Conclusions. In the course of our work, we obtained a complete neural network that allows us to identify the five personal qualities and the type of character of an athlete based on his face photo. We tested the percentage of correct answers obtained with different numbers of possible epochs implemented by us: we found out that with 30 epochs of training, the highest percentage is 29.5%, while the athletes themselves are accurate only in 25% of cases.