Forecasting Methods for Horseracing

Author: Peter May
Publisher: High Stakes
Buy this book at Amazon

<< Previous Page   Next Page >>

Peter May

Another advantage with feature extraction is the reduction in dimensionality which normally accompanies such an operation. For instance, in the character classification problem discussed above 100 inputs would be required to represent the raw data. As the number of dimensions increases the volume of historical data required to build the model also increases. Increasing dimensionality rapidly leads to sparse data sets and, as a result, poor input-output mappings. For the horseracing problem a great deal of pre-processing is required. The majority of the data is included in race results and hence needs conversion into an acceptable form. Intermediate conclusions about the animal's likes and dislikes (i.e. suitable race distance) can be gleaned from these results and used as inputs to the model as opposed to the complete race result. Furthermore, it is possible to include prior knowledge in the form of the indicators to the general level of ability of the horse (i.e. a rating) or previous success rates of the trainer or jockey (i.e. classifying the experienced jockeys from the less experienced riders). Methods for pre-processing the horseracing data using feature extraction, inclusion of prior knowledge, and techniques for handling missing data are discussed in Chapter 3.

_______________________ Objectives

The primary aim of this book is to demonstrate how techniques taken from the field of artificial intelligence can be used to generate forecasting methods for the horseracing problem. Although three simple rule-based systems that identify good betting opportunities are presented in Chapter 5, the main focus of this text concerns more complex, computer-based, forecasting methods. Consequently, a knowledge of computer programming and access to a computer is considered beneficial, although a version of the knowledge-based approach could be implemented using a spreadsheet Apart from the rule-based systems, this text does not include ready-to-use methods, instead it provides a discussion of techniques which are applicable to the horseracing domain and illustrates how these techniques can be combined into forecasting

18