Combining Deep Learning and Probabilistic Model Checking in Sports Analytics

Kan Jiang

Abstract: Deep Learning (DL) is good at finding the patterns hidden in big data, while Markov Decision Process (MDP) is good at modeling the dynamics in a complex system for formal analysis, e.g. Probabilistic Model Checking (PMC). The two models complement each other. Unlike the black box DL-Only model, the combined model is interpretable. Unlike the MDP-Only model, the combined model is able to draw deep insights from the data. Both interpretability and capability of finding deep insights are desirable in many applications, including sports analytics. In this paper, we propose to combine DL and PMC, and apply it in sports analytics to find an accurate and interpretable winning strategy.