A new article from Sports Trader concerns Adaptive Boosting and provides plenty of detail and references for those wanting to learn more about this particular form of machine learning.
My only query is the application, determining the winner of a horse race. Any approach to this problem falls at the first hurdle because it usually lacks private information.
Any machine learning methodology can be used to trawl through public data in a reasonable length of time - also brute force search - to determine fair odds. See Betfair Trading Techniques.
Without private information or the ability to act upon late-breaking information more quickly than other traders, filter rules based on public data are unlikely to give a significant long-term edge.
This is why I base my research around taking advantage of novice traders and others who overreact or act too late in dynamic markets.