Prediction model of basketball players' playing time based on neural network

Abstract

The purpose of this study is to predict the playing time of CBA league players through neural network model, and to explore the key factors affecting the playing time from the perspective of quantitative analysis, so as to provide data support for coaches to make decisions on arranging players to play. This paper selects 7340 items of average data of 367 players in CBA league in the regular season of 2021-2022 as the research object. In model training, other data indexes except playing time are used as input parameters, playing time is used as output variable, and automatic encoder is added to screen key data indexes, thus establishing playing time prediction model. The results show that five models and a total data model are established according to the players' positions on the field (point guard, shooting guard, small forward, power forward and center), and the highest value of the average error (MER) is 1.56 and the lowest value is 1.42. R2 is 0.785 at the highest and 0.726 at the lowest. The results show that the data indexes that affect playing time are position-specific, and the models established for different positions have high prediction ability for players' playing time. The average error of the total data model is the best, while the explanatory ability (R2) of the small forward model data is the best, which proves that each model can provide data support for coaches' decision-making.