A Comparative Study Of Supervised Machine Learning Models For Predicting Bowler Performance In T-20I Cricket

Abstract

This study investigates the performance of the top 100 T20 International (T20I) bowlers, utilizing data sourced from Cricinfo to analyze key performance metrics that influence bowlers' success in the format. The research employs various classification algorithms, including Decision Trees, Naïve Bayes, Logistic Regression, Support Vector Machines, Extreme Gradient Boosting, and Random Forests, to categorize bowlers based on attributes such as age, bowling style, and playing role. Data was collected through web scraping techniques, focusing on match statistics and performance-specific metrics. Results indicate that the Decision Tree classifier achieved the highest accuracy (85%) in classifying bowlers into spin and fast categories, while Random Forest exhibited lower performance (60%). The study highlights the significance of age, bowling style, and performance metrics in determining bowler classification and effectiveness, emphasizing the need for further optimization and feature engineering in the predictive modeling of bowler performance.