A feature engineering approach to predicting player performance in basketball

Recent advancements in sports analytics have found many fields of applications in basketball. Player performance prediction is one of the main goals of basketball analytics because of the potential implications for both teams and fans. This study aims to create a predictive modeling approach that is designed for accurately estimating the performances of basketball players while addressing the main issues in basketball statistics. Euroleague, the highest level of European basketball club competition, is selected to conduct our study. The data set used in this study contains 720 regular- season games from 2016-2017, 2017-2018, 2018-2019 Euroleague seasons. During these seasons, a total of 15368 records obtained from performances of 464 individual athletes. In order to create models for predicting performances of basketball players, we followed a structured data mining process. Most predictive models in the literature have relied on offensive statistics because of the scarcity of statistics that are related to defense. However, this study addresses the need for defensive metrics in player performance prediction, so far lacking in the literature. We developed a methodology and proposed a feature engineering approach to create data-driven defensive metrics. Our results demonstrate that the most significant boost in both R-squared and rmse values have been achieved after adding position-based defensive metrics.