Análise de Atributos e Previsão de Desempenho de Jogadores Novatos na NBA Utilizando Métodos de Boosting

Abstract

The use of machine learning techniques in the sports industry brings several benefits to the sport, one of which is aiding in the identification and discovery of the best attributes to determine good players. This study addresses the use of machine learning techniques in the sports context, specifically in the NBA, with the goal of identifying players with above-average Win Shares per 48 (WS/48) and determining the most important attri butes. The justification for this study lies in the importance of player performance in the NBA and the need to identify which characteristics are more relevant for the success of players who participated in the draft. Utilizing data from college basketball and the NBA, five boosting-based classification methods were tested, with CatBoost achieving the best results, reaching 92% accuracy. The most relevant attributes for classification were points per game, rebounds per game, and assists per game. This study contributes to understanding the factors that influence player performance in the NBA, providing valuable insights for teams and coaches in player selection and development. Future work may explore the exclusive use of college data to predict player performance prior to their selection.