In-game winner prediction and winning strategy generation in cricket: A machine learning approach

Abstract

This study provides an in-game prediction of the winner for Twenty20 (T20) cricket by focusing on the matches played in the Indian Premier League. For the analysis, data were collected from 812 completed matches played between 2008 and 2020. Initially, several candidate features were identified, and then the LASSO method was applied as a feature selection technique to identify the most important set of features. Based on the identified important features, predictions are provided for each stage of a match where a T20 match can consist of a maximum of 240 stages. For each stage, three classification models were formed using Naive Bayes, Logistic Regression and Support Vector Machines. The prediction accuracy was used for evaluating the findings of the study, and the prediction accuracy of each model indicates the ratio between the number of correctly predicted instances and the total number of predicted instances. Naive Bayes demonstrated prediction accuracies ranging from 53.08% to 91.76% between the first and the 240th stage of matches, whereas the accuracy of Logistic Regression varied from 56.92% to 97.65%. In comparison, Support Vector Machines also displayed comparable outcomes with a prediction accuracy of 55.00% at the first stage, and 90.59% at the 240th stage. Furthermore, a strategy generator that assists the competing teams in the second innings to devise winning strategies, is presented in this study alongside an interactive web-based application for making in-game predictions, and for assisting the end users (players and coaching/managing staff) in decision making, based on the generated winning strategies.

RunsGuard Framework: Context Aware Cricket Game Strategy for Field Placement and Score Containment

“…Mahbub et al [30] and Pussella et al [31] focused on cricket match forecasting. The most used techniques for this study were the Support Vector Machine (SVM), Naive Bayes, and Random Forest (RF) in machine learning to predict the top 11 cricket players for Bangladesh in One-Day International (ODI) matches.…”