The Equine Forecast Model was developed using thousands of thoroughbred races from multiple tracks, distances and racing surfaces over a 3-year period. Data inputs were built from race entries and associated past performances for each horse to create a large data-mart of model observations. There were hundreds of different data elements for each race/horse entry.
The initial data exploration phase revealed that different types of races are affected by combinations of different factors. Races were grouped together in segments based on a combination of distance, race surface and class; individual component models were developed for each segment to optimize the forecast. (The combined model is known as an ensemble model.)
After the model evaluation phase, the final list of significant model “features” uses a number of derived elements including early/late pace values, normalized speed ratings (adjusted for different tracks, surfaces and distances) and current horse “condition” rating.
To actually use the model, new racing data can be downloaded and processed using the model coefficients to produce new forecasts. A detailed race-by-race list of contenders is produced with past races and overall performance projections.
The intent of the forecast is not to produce an iron-clad list for betting, but to rank the contenders which then should be individually checked for earlier performances at the same distance, surface and class.
Following the race entries, a category summary is available to help align the bettor’s personal approach (speed, pace, class, workouts, jockey) with each contender.
