Integrating machine learning and decision support in tactical decision-making in rugby union

Rugby union, like many sports, is based around sequences of play, yet this sequential nature is often overlooked, for example in analyses that aggregate performance measures over a fixed time interval. We use recent developments in convolutional and recurrent neural networks to predict the outcomes of sequences of play, based on the ordered sequence of actions they contain and where on the field these actions occur. The outcomes considered are gaining territory, retaining possession, scoring a try, and being awarded or conceding a penalty. We consider several artificial neural network architectures and compare their performance against baseline models. Accounting for sequential data and using field location improved classification accuracy over the baseline for some outcomes. We then investigate how these prediction models can provide tactical decision support to coaches. We demonstrate that tactical insight can be gained by conducting scenario analyses with data visualisations to investigate which strategies yield the highest probability of achieving the desired outcome.

Automated summary

Research utilizing neural networks to predict rugby match outcomes and provide tactical guidance found that considering sequential data and field location can improve classification accuracy. Conducting scenario analyses with data visualizations can provide tactical insights on which strategies are most likely to achieve the desired outcome.