Horse Racing Prediction Using Artificial Neural Networks

Artificial Neural Networks (ANNs) have been applied to predict many complex problems. In this paper ANNs are applied to horse racing prediction. We employed Back-Propagation, Back-Propagation with Momentum, Quasi-Newton, Levenberg-Marquardt and Conjugate Gradient Descent learning algorithms for real horse racing data and the performances of five supervised NN algorithms were analyzed. Data collected from AQUEDUCT Race Track in NY, include 100 actual races from 1 January to 29 January 2010. The experimental results demonstrate that NNs are appropriate methods in horse racing prediction context. Results show that BP algorithm performs slightly better than other algorithms but it needs a longer training time and more parameter selection. Furthermore, LM algorithm is the fastest.