SPORTS STRATEGY ANALYTICS USING PROBABILISTIC MODEL CHECKING AND MACHINE LEARNING

Abstract

This multi-disciplinary research work applies formal methods, machine learning, and computer vision to a novel application domain of sports strategy analytics. We present the use of probabilistic model checking to predict the winning chance of a sports player based on his play pattern and reliabilities of sub-skill set, and how to recommend effective strategies to improve his winning chance. To facilitate the strategy analysis, we propose deep learning-based multiple models with computer vision algorithms to extract action sequence data from YouTube videos. This task is non-trivial as ball tracking and action identification from low-resolution video provide many technical challenges. We further extend our approach to team sports, using an approximated value iteration algorithm combined with Adaptive Q-value Tree Search to approximate winning chance accurately. This new method is generic and can be applied beyond sports analytics into other domains.