Framework for Analysis and Prediction of NBA Basketball Plays: On-Ball Screens

In this study, we present a framework for automatic identification and analysis to predict the NBA offensive basketball play: On-Ball Screen. In basketball, the on-ball screen is a dynamic offensive strategy that involves the movement of multiple players and the ball to create an effective shot attempt. It is a fundamental play employed by all the teams in the National Basketball Association (NBA), the highest level of competition for basketball. With the presence of sophisticated data collection and analysis tools, this paper presents methodologies to process spatial and temporal data and big data in complex multimedia forms for sports analytics. We propose a framework to extract, transform, and analyze player motion-tracking data in NBA games and apply machine learning techniques with improved feature selection methods to predict the presence of on-ball screens. Results show an area under receiver operating characteristic (AUROC) of 0.9552, 0.13 increase and 4% Accuracy improvement from existing literature.