Optimal Machine Learning Models for T20 Cricket: The Role of Dangerous Balls in Match Outcomes

Abstract

This study explores the application of machine learning models Logistic Regression, Multilayer Perceptron (MLP), and Decision Tree (CRT method) to predict the outcome of T20 International cricket matches based on dangerous deliveries consisting on two key independent variables: wickets lost and extras conceded. The dataset, comprising 2,492 matches, was analyzed to understand the impact of these variables on match results. By comparing the performance of these models across the first and second innings of matches, the study aims to identify which model best captures the dynamics of match outcomes. The models were evaluated in terms of their predictive accuracy, interpretability, and the significance of the variables, with a particular focus on the role of wickets in determining match results. Although all three models proved effective in predicting match outcomes, the Decision Tree model stood out as the most reliable and comprehensible, providing meaningful insights into the connection between match dynamics and results. The findings highlight the potential of machine learning techniques in sports analytics, offering valuable insights for both researchers and cricket analysts in forecasting match outcomes and understanding the factors that influence a team's success in T20 cricket.