Winner prediction in an ongoing one day international cricket match

Abstract

Cricket is a team sport with an intricate set of rules, where players specialize in multiple skills such as batting, bowling, and fielding. Playing conditions and home advantage also impact the game. Thus, it is quite challenging to build an accurate quantitative model for the game. In this paper, we provide a data driven approach to predict the winner of a cricket match. We divide the ongoing match into various states and provide a prediction for each state using supervised machine learning models. We employ dynamic features that account for the current match situation, together with the static features like team strength, winner of the toss, and the home advantage. We also use SHAP scores—an explainable AI technique—to interpret the proposed prediction model. We use ball-by-ball data from 1359 men’s one day international cricket matches played between January 2004 to January 2022 to present our results. We achieved the best in-play prediction accuracy of about 85% . SHAP scores reveal that during initial phases of the match, the model treats static features like team strength more important than others, in making the predictions. But as the match progresses, dynamic features capturing the current match situation become exceedingly important. Our work may be useful in preparing tools for in-play winner prediction for live cricket matches that can be used in websites and mobile applications covering the sport, in providing analytics during live television commentary, and in legal betting platforms.