Beyond the run-rate: forecasting framework for first innings score in t20 cricket

Abstract

With the popularity of the T20 cricket format, the game of cricket has dramatically changed compared to several decades ago. Every year there are more than 100s matches played, which results in thousands of data that can be used by sports analysts in cricket. Several studies have attempted various analyses of the game, such as predicting the likelihood of a team’s victory, analyzing individual player performances and forecasting scores. However, forecasting scores has not been studied extensively and limited to specific teams, rather than a generalized approach. Our paper presents a generalised novel deep neural network-based method to predict the score of the first innings in a T20 international cricket match. The model utilizes various attributes in three categories namely a) current status of the match b) performance of the current batsmen and c) performance of the bowler and provides predictions for each over. We have used recent 5 years T20 international matches from 14 teams and tested our method in the 2022 ICC Men’s T20 World Cup. We demonstrate our findings quantitatively and qualitatively in this paper.