Predicting English Premier League Matches Using Machine Learning and Conditional Probability

Abstract

This research addresses the continuous challenge of accurately predicting football match outcomes, which is crucial for the sports betting sector's profitability and reliability. Considering the complex nature of factors influencing results, the study employs a quantitative approach and integrates Bayes' conditional probability theory using Gaussian Naïve Bayes. By leveraging English Premier League data from 11 complete seasons and half a season, the developed model achieves a training accuracy of 87% and an average testing accuracy of 85%. Comparative analysis with existing studies reveals competitive performance, albeit trailing certain advanced models. Despite the need for further refinement, the model offers a profitable avenue for betting markets, emphasizing the importance of ongoing enhancements in feature engineering. Overall, this research contributes to the field by providing a robust predictive model with potential implications for both bookmakers and punters.