Probability prediction using support vector machines

Reports on an investigation of the use of support vector machines in the problem of beating bookmakers´ odds. The method used is to create nine regression lines, one for each of the possible half-time/full-time possibilities, e.g. a draw at half time with the home team winning at full time. We extend the support vector machine by incorporating variable-width permissible error bounds based on our knowledge of the sample size on which each of the data points is based. We show that the variable-width method improves the regression