Data-Driven Analytics for Decision Making in Game Sports

Performance analysis and good decision making in sports is important to maximize chances of winning. Over the last years the amount and quality of data which is available for the analysis has increased enormously due to technical developments like, e.g., of sensor technologies or computer vision technology. However, the data-driven analysis of athletes and team performances is very demanding. One reason is the so called semantic gap of sports analytics. This means that the concepts of coaches are seldomly represented in the data for the analysis. Furthermore, sports in general and game sports in particular present a huge challenge due to its dynamic characteristics and the multi-factorial influences on an athlete’s performance like, e.g., the numerous interaction processes during a match. This requires different types of analyses like, e.g., qualitative analyses and thus anecdotal descriptions of performances up to quantitative analyses with which performances can be described through statistics and indicators. Additionally, coaches and analysts have to work under an enormous time pressure and decisions have to be made very quickly.
In order to facilitate the demanding task of game sports analysts and coaches we present a generic approach how to conceptualize and design a Data Analytics System (DAS) for an efficient support of the decision making processes in practice. We first introduce a theoretical model and present a way how to bridge the semantic gap of sports analytics. This ensures that DASs will provide relevant information for the decision makers. Moreover, we show that DASs need to combine qualitative and quantitative analyses as well as visualizations. Additionally, we introduce different query types which are required for a holistic retrieval of sports data. We furthermore show a model for the user-centered planning and designing of the User Experience (UX) of a DAS.
Having introduced the theoretical basis we present SportSense, a DAS to support decision making in game sports. Its generic architecture allows a fast adaptation to the individual characteristics and requirements of different game sports. SportSense is novel with respect to the fact that it unites raw data, event data, and video data. Furthermore, it supports different query types including an intuitive sketch-based retrieval and seamlessly combines qualitative and quantitative analyses as well as several data visualization options. Moreover, we present the two applications SportSense Football and SportSense Ice Hockey which contain sport-specific concepts and cover (high-level) tactical analyses.