A case study using neural networks algorithms: horse racing predictions in Jamaica

Abstract

Neural networks (NNs) have been applied to predict many complex problems, such as horse racing prediction. This study investigates the performances of four supervised neural network algorithms in horse racing. We employed the Backpropagation (BP), Quasi_Newton (QN), LevenbergMarquardt (LM) and Conjugate Gradient Descent (CGD) learning algorithms to real horse racing data collected from Caymans Race Track, Jamaica. The training and testing data comes from 143 actual races collected from 1 January to 16 June 2007. The experimental results demonstrate that all algorithms can provide acceptable predictions with an accuracy of 74%. The BP algorithm slightly performs better than the other three algorithms but needs a longer training time and more parameter selections.