Development and application of computer-based prediction methods

Abstract

The purpose of the current study was to develop an information database which accounted for as many of those variables thought to cause variance in soccer and rugby union performance as possible. These data were used to populate seven different predictive models of performance which were compared retrospectively to actual results. A total of 7 independent variables were applied to a match database 3 seasons in length for 3 English Premiership Football teams and 2 Premiership Rugby Union teams. Prediction methods included Multiple Linear Regression, Artificial Neural Networks and expert human predictions.Contrary to previous literature, soccer (57.9%) was on average predicted more successfully than Rugby Union (46.1%). Nevertheless results suggested that the ability of Artificial Intelligence and Computerised methods to predict the outcome of matches has, for the first time, surpassed that of humans.