Trading "in-play" betting Exchange Markets with Artificial Neural Networks

Abstract

In this thesis I have studied, analyzed and implemented a solution for using artificial neural networks for prediction of in-play tennis match odds markets on Betfair. The overall prediction task was concentrated on maximizing potential profit, rather than just minimizing some standard error. The properties of odds trading were studied, and on the basis of this, a new cost function suitable to the underlying problem was proposed. The new cost function tried to capture some of the problem specific characteristics, and aimed to maximize the return of each trade. Training with both the new and the standard cost function was conducted, together with a range of training algorithms, ANN structures and parameters