Performance Analysis of Classification Algorithms for Outcome Prediction of T20 Cricket Tournament Matches

Abstract

Cricket is the most popular sport in south Asian countries and the second most popular sport globally. And T20 cricket is the most attractive version of this game. Businesses have grown enormously based on cricketing sports events from the last decade. A large number of research have been done to predict the winner of cricket matches or to analyze game statistics. These studies are helping investors and franchisees to decide on which team they can invest in to gain more profit. Also, coaches, sports analysts, and technicians get game facts and ideas about other teams, which help them make decisions and change plans accordingly. In this study, we have shown a comparative analysis of different non-ensemble and ensemble machine learning classifiers. We have collected data of all sea-sons of the Bangladesh Premier League(BPL) T20 tournament. We have prepared two datasets: one contains only pre-match information, and the second includes post-match information. We have applied five base classifiers and five ensemble classifiers for the analysis. We found that K-Nearest Neighbor(KNN) performed best compared to other algorithms while predicting match results before starting the game. The Gradient Boosting classifier seemed more robust than the different classifiers for predicting match outcomes considering all features.