How our Horse Racing Predictions are Made
Since early 2001 we have been using the 3rd version of our AI horse racing system. We now use the excellent open source machine learning toolset from H2O.
The first version of our free AI horse racing predictions used a bespoke artificial neural network written in Java. Although converging well the solution was extremely slow, taking days to train on a single course dataset.
We then upgraded to use FANN - Fast Artificial Neural Network written in C (a proper programming language).
Domain knowledge is vital for any AI system, although passionate followers of horse racing we required expert insights from within the industry to point us in the right direction for which factors affect the results. Using FANN we were then able to test multiple different combinations of these variables which we now use with H2O.
The goal is to automate the process of generating and publishing the tips on the website. The backbone of the system is written in Perl (yes, I'm old school). Training the model and creating the dataset is still a highly manually and time consuming effort, hence the long periods when the website has not been updated.
We now feel we are generating our best predictions, but whilst AI can identify patterns and provide valuable insight into Horse Racing it is important to note that these tips cannot be 100% accurate. Therefore, our horse racing predictions should only be used in conjunction with other forms of analysis, other sources of tips and expert opinion.