Predicting rugby world cup outcomes with machine learning: A comparative model evaluation

Predicting the outcome of rugby union matches has traditionally relied on expert judgement and historical observation, an approach that is slow, difficult to scale across large fixture lists, and susceptible to human bias. This study investigates whether supervised machine learning can provide a more objective and scalable alternative, using the 2023 Rugby World Cup qualification-era fixtures as the evaluation context. A dataset of international rugby results spanning 1871 to 2024 was compiled, cleaned, and enriched through feature engineering, including team ranking points, recent-form indices, and match-experience counts. Four classification algorithms; Logistic Regression, Support Vector Machine with a radial basis function kernel, Random Forest, and AdaBoost, were trained on an 80/20 train-test split and evaluated using accuracy, precision, recall, F1-score, and area under the ROC curve (AUC), with hyperparameters tuned via grid and randomised search. Random Forest achieved the highest overall accuracy (72%) and the strongest discrimination for home victories (AUC = 0.88), followed by AdaBoost (70%), and Logistic Regression and SVM (69% each). All models struggled to predict draws, a consequence of severe class imbalance in the underlying data. The findings confirm that ensemble tree-based methods offer a modest but consistent advantage over linear and margin-based classifiers for rugby outcome prediction, while highlighting class imbalance and limited feature richness as the principal barriers to further improvement. The paper concludes with recommendations for resampling strategies, external validation, and the inclusion of player- and weather-level features in future work.