Cricket Winner Prediction - Domain-Based Feature Engineering and Analysis
University of Dhaka, Dhaka, Bangladesh, 2023
Cricket, one of the world’s most captivating sports, presents a unique challenge in predicting match outcomes. This study delves into the complexities of forecasting winners by analyzing a combination of external and internal factors related to the teams. Through domain-based feature engineering, we create new predictors to enhance accuracy, uncovering the intricate dynamics of the sport.
Our research utilizes datasets from Kaggle and Cricsheet, incorporating features like expected runs, economy rates, and home advantage indicators. Using four machine learning models—Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine (SVM)—we achieved up to 72% accuracy with SVM. However, a low correlation between features and match outcomes suggests significant potential for future improvements.
This study highlights the evolving intersection of sports and data analytics, offering a foundation for further exploration in cricket match prediction. By integrating advanced techniques like time-series analysis or deep learning and incorporating weather data, the predictive models could achieve greater precision and reliability.