This is a complex horse racing system and likely to require the assistance of computer based software. However I do know some people who have been successful with a simpler version of this system that does not require dedicated software.

The system is based on the theory that certain factors (weightings) should be applied to 'form variables' (weight, track performance, barrier, distance performance etc) to predict the outcome of a race. The trick is to determine which form variables to factor and by how much.

For example: If I have 4 horses in a race (to make it simple) and using the form guide I rate form over a distance as follows:

HorseA=80
HorseB=90
HorseC=60
HorseD=100

(don't pay too much attention to these numbers its a score I apply based on their form over the distance)

How much do these scores contribute to the horse's total score for the race e.g if it is half (0.5) the final score for this 'form variable' would be

HorseA = 40 (0.5 * 80)
Horse B = 45 (0.5 * 90) etc

My approach to this challenge is to calculate and record the form variables for each race (mainly weekend races at Metropolitan racecourses) along with the results for each race. Results include placings from first to last including the lengths of separation. I feed this information into the computer software to determine what factors would have to be applied to the form variables to provide me with the correct placings in the race. There is usually many successful factor combinations for each race.

The computer software then analyses these factor combinations against successful combinations from other races and attempts to identify patterns (trends) amongst the form variables. For example at certain tracks, with heavier going certain barriers tend to produce more winners.

The key challenge with this approach is when a horse faces an extraordinary circumstance in a race, which caused them to finish back in the field. Initially I did not concern myself with these events. However now I record a seperate set of history which omitts these horses from the selection analysis and the results (ie the learning system ignored they ever ran). As a result the predictions are slightly more consistent.

I hope this has not confused you too much and is of some use.