Predicting English Premier League Final Rankings Using Contextual Machine Learning and Monte Carlo Simulation

Abstract: Forecasting the final English Premier League (EPL) table is challenging due to football’s low-scoring nature, outcome variability, and evolving team strength during the season. This research presents a pipeline that predicts match outcomes using contextual features and simulates the entire season to generate probabilistic final rankings, following the season-simulation tradition in football forecasting. Historical EPL match data from 2019 to 2024 are used to train probabilistic classifiers, specifically Random Forest and Gradient Boosting, implemented using scikit-learn, to estimate the probabilities of home win, draw, and away win. For the 2025–26 forecast, the model relies solely on the published fixture list and pre-season team context, including summer transfer-window net spending, without incorporating any match results from the target season. Match probabilities are converted into a distribution over final tables using Monte Carlo. The results include the average and median finishing positions for each club, as well as the probabilities of finishing in the top four and of relegation.