Regression to Forecast: An In-Play Outcome Prediction for One-Day Cricket Matches

The paper presents a technique for predicting the outcome of an ongoing cricket match. A robust prediction is obtained by taking into account the quality of players and form of the team in addition to the toss, host advantage, and day/night match facts. A novel approach is used to quantify the player quality and form of the team. At each stage, the explanatory power of the available parameters is analysed to include in the model, replacing few existing parameters, for effective prediction. The quantification of these parameters is subjected to dynamic logistic regression, as the parameters change as the match progresses. The model is evaluated against 187 ODI matches between 2015 and 2017, and a sound prediction accuracy is observed. The model is evaluated in five different scenarios of the match as follows: before match begins, after 30 overs in first innings, at the end of first innings, and after 30 and 40 overs in the second innings.