Bangladesh Premier League T20 Cricket Match Outcome Prediction Using an Ensemble Learning Approach

Predicting the results of cricket matches, especially in the fast-paced Twenty20 (T20) format, is a significant challenge in sports analytics. This paper presents a comparative study of machine learning models for predicting match outcomes in the Bangladesh Premier League (BPL). We evaluate the performance of three advanced ensemble models-Random Forest, Gradient Boosting, and XGBoost-against a single Decision Tree baseline. Using a comprehensive historical dataset spanning from the 2012 to 2024 BPL seasons, our approach incorporates richly engineered features, including team form metrics and venue-specific dynamics, alongside core match attributes. The models were rigorously evaluated using 5-fold cross-validation. The results demonstrate the superiority of ensemble techniques, with Gradient Boosting and XGBoost achieving the highest mean accuracies of 97.73 % and 97.70%, respectively, significantly outperforming the baseline model. These findings underscore that a combination of sophisticated feature engineering and advanced ensemble algorithms is crucial for capturing the complex, non-linear dynamics of T20 cricket. The framework provides a valuable tool for teams, analysts, and fans, laying the groundwork for more accurate predictive systems in the future.