Soccer Strategy Analytics Using Probabilistic Model Checkers

Rajdeep Singh Hundal, Zhaoyu Liu, Bimlesh Wadhwa, Zhe Hou, Kan Jiang, Jin Song Dong

Abstract: When it comes to predicting the outcome of soccer matches, there are two main techniques that dominate the betting industry: deep learning and machine learning. However, as with many state-of-the-art applications, these methods often exhibit significant drawbacks. Primarily, they act as black boxes where the internal operations are not transparent, making it difficult to discern the cause when predictions fail. In this paper, we explore the potential of using probabilistic model checkers as an alternative approach for predicting soccer match outcomes. This technique, while unconventional, offers greater transparency or white box visibility into its operations when compared to state-of-the-art methods. The choice of utilizing probabilistic model checkers is often overlooked due to their propensity to induce state explosion in complex models. To address this, we propose specific strategies aimed at minimizing the state space of a soccer match, thereby mitigating the state explosion issue. We assess the effectiveness of the probabilistic model through various metrics and compare its performance against established deep learning and machine learning baselines. To evaluate its real-world applicability, we also simulate betting based on the predictions of the probabilistic model. This paper concludes by addressing the practical challenges involved in implementing such a predictive model.

External IDs:dblp:conf/isace/HundalLWHJD24