Filtering active moments in basketball games using data from players tracking systems

In recent years, sport analytics evolved in the massive collection of data, especially from Global Positioning System (GPS) sensors installed in sport facilities or worn by the athletes. The largest amount of data are used to track locations and trajectories of players during their performance. Data analysis of positioning information during the actions of a game allows a deep characterization of the performance of single players and the whole team. Basketball is one of the team sports where analytics are becoming a fundamental asset. However, during a game, actions are interleaved with inactive periods (e.g., pauses or breaks). For a proper knowledge extraction on the game features, the analysis of players movements must be restricted to active periods only. This paper proposes an algorithm to automatically identify active periods by using players’ tracking data in basketball. The algorithm is based on thresholds that apply to players kinematic parameters. The values of thresholds are identified by setting-up a “ground truth” extracted from the video analysis of the games and by developing a performance evaluation method derived from “Receiver Operating Characteristic” (ROC) curves. When tested on a number of real games, the method shows good performance. This algorithm, along with the identified parameters, could be adopted by practitioners to identify game active periods without the need for video analysis.