Predictive Modeling of Wicket and Extras Occurrence within the Over Using Neural Network

Abstract

This study investigates the use of Multi-Layer Perceptrons (MLPs) to predict ball-specific outcomes in cricket matches, focusing on wicket-taking deliveries and extras across the six balls in an over. The data, sourced from reputable cricket databases such as ESPN Cricinfo and Cricsheet, revealed key patterns of wicket and extra distributions, with Ball 5 showing the highest proportion of wickets and Ball 4 the highest proportion of extras. However, the MLP model, despite its potential to capture non-linear relationships, demonstrated significant challenges in achieving high predictive accuracy. While it performed best for Class 6, the overall accuracy remained low, with poor performance observed across most classes, indicating issues like class imbalance and insufficient feature representation. The model’s discriminative power was limited, as reflected in the ROC curves and cumulative gain and lift charts, suggesting a need for improvements in model architecture and feature engineering. The study highlights the importance of integrating ball-specific patterns into predictive models for cricket match outcomes, and suggests that exploring alternative machine learning algorithms, such as Random Forests or XGBoost, could lead to better prediction accuracy. These findings provide valuable insights into improving the predictive capabilities of cricket data analysis models, particularly by addressing the underlying challenges in classifying ball-specific events.