A0264
Title: Lasso multinomial performance indicators for in-play basketball data
Authors: Argyro Damoulaki - Athens University Economic Business (Greece) [presenting]
Ioannis Ntzoufras - AUEB (Greece)
Konstantinos Pelechrinis - University of Pittsburgh (United States)
Abstract: A typical approach to quantify the contribution of each player in basketball uses the plus/minus approach. Such plus/minus ratings are estimated using simple regression models and their regularized variants with the response variable, either the points scored or the point differences. To capture more precisely the effect of each player and consider specific lineups, play-by-play data are needed. The aim is to investigate the performance of regularized adjusted plus/minus (RAPM) indicators, estimated by different regularized models having as a response the points scored per possession, using 2021-2022 NBA possession data (n=322,852). Simple model-based indices are initially presented starting from the ridge regression, the standard technique in the relevant literature. The lasso approach is proceeded with, which has specific advantages and better performance than ridge when compared with selected objective validation criteria. Then, regularized logistic regression models are implemented to obtain more accurate indicators since the response is a discrete variable, taking values mainly from zero to three. The final proposal is an improved RAPM measure, which is based on the expected points of a multinomial logistic regression model where each player's contribution is weighted by his participation in the team's possessions. The proposed indicator, called weighted expected points (wEPTS), outperforms all other RAPM measures we investigate.
