T20 cricket match score and winning team prediction using machine learning techniques

Abstract

Cricket is one of the famous outdoor sports that contain an outsized set of statistical data in the world. As T20 games rise in popularity, it's necessary to look at the possible predictors that affect the result of the matches. This research aims at analyzing the T20 cricket match results from the dataset collected (2005-2021). It focuses on measuring the result of T20 matches by applying the prevailing machine algorithms learning to the balanced also as an imbalanced dataset. The database used here was an unbalanced database and had to be converted into a balanced database. Therefore, this research was performed on this unbalanced dataset as well as the balanced dataset. Oversampling technique is employed for imbalanced datasets then the algorithm is applied. Accuracy is used as the performance metric and calculated by using machine learning algorithms. It is also considered as evaluation criteria and the percentage will vary consistent with the various algorithms. Three models were created basically, and the third model was a hybrid model created with the output of the first and second models. In the first model, the Random Forest algorithm obtained an accuracy of 84.51% for the imbalance data and 76.61% for the balance data. The Decision Tree algorithm was used to create the third hybrid model, with an accuracy of 97.23% for the imbalanced data and 93.92% for the balanced database. Thus, the hybrid model operates with higher accuracy than the first model. The first model is designed to predict the winning team before the start of the game, the second model is used to predict the score of the team batting in an innings, and the third model is used to predict the winning team in the second inning.