The application of self-organising maps to performance analysis data in rugby union

With the advent of professionalism in rugby union greater volumes of information are collected about player and team performance. A typical Opta^TM, Sports Code^TM timeline, for a single rugby match, can have more than 2000 instances and labels of information. Unless there is a prior understanding of an opponent much time can be spent identifying irrelevant trends and information which may not fairly represent the performance of the match.

Kohonen Self-organising Maps (SOMs) are a form of artificial neural network developed for clustering and visualising high-dimensional data by reducing the output to a low-dimensional output map. These visualisations may help the analyst quickly identify important relationships among the key performance indicators describing a match.

In this paper we report on the application of SOMs to discrete data summarising matches in New Zealand's ITM Cup rugby competition. The input variables were frequencies of common performance indicators. The SOM approach was used to narrow down the input variables to those that discriminate between successful and unsuccessful outcomes as well as map regions associated with various levels of success. Since map regions indicate game patterns or styles, further analysis showed that multiple game styles tended to lead to wins and multiple different styles tended to lead to losses. SOMs represent an important method for characterising game play in rugby union, we suggest continued use of SOMs will help make coaches and analysts more familiar with their interpretation and anticipate further streamlining of key performance indicator selection.