Developing a High-Performance Sports Results Prediction Artificial Neural Network: Case Study on World Championship Boxing

Abstract

Abstract Major sports events are watched by millions around the world and the prediction of event outcomes is a subject of interest to many stakeholders which underlines the relevance of continuous development and improvement of prediction models. This study uses a factorial design methodology to develop and test 18 Artificial Neural Network (ANN) models for the prediction of world championship boxing matches. The methodology was applied to evaluate the individual and collaborative effects of feature selection, ANN architecture and training data selection on the prediction performance of ANNs. Feature selection was found to be the most influential factor on prediction performance with a statistically significant Analysis of Variance (ANOVA) between the feature selection levels and the test accuracy (p-value of 0.012). The collaborative effect of training data selection and feature selection on prediction performance was found to be statistically significant with ANOVA p-value of 0.007. The best performing model achieved a test accuracy of 81.53% which is an improvement to current benchmarks for sports prediction. The findings of this study contribute to the development of future machine learning sports prediction models.