Using Data Mining to Create Meaningful Rugby Ratings for Individual Performance
by
Paul Bracewell
Massey University, Albany

The flexibility, variety and availability of data mining software enables the extraction of latent information relating to sporting performance on a match-by-match basis from multivariate data. However, for these individual performance ratings to be meaningful, the underlying model must use suitable data, and the end model itself must be contextual, robust and transparent.

Three 'unsupervised' data mining techniques - factor analysis, self-supervising feed-forward neural networks and self-organising maps - are applied to estimate individual rugby player performance using data provided by Eagle Sports (www.eaglesports.co.nz). For each method, the problems of context, robustness and transparency in reduced dimensionality are discussed along with possible solutions. Of particular importance is the introduction of the half-moon statistic, which is a new functionally independent test for independence between variables.

The resolution of these problem areas allows meaningful ratings for individual rugby player performance to be extracted from the Eagle Sport database. These ratings are useful tools for diagnostic coaching; enabling coaches, selectors and other interested parties to identify strengths and weaknesses in player performance. This is achieved by deconstructing individual performances to identify the skill set(s) responsible for aberrant performances.

Date received: May 7, 2002

Copyright © 2002 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cajj-05.