California Scientific

California Scientific
1000 SW Powell Ct
Oak Grove, MO 64075
Sales@CalSci.com
916-225-5119
California Scientific  *  BrainMaker Neural Network Software  *  Predict Forecast Classify Stocks Bonds Markets Commodities Diagnose Medical

BrainMaker Neural Network Software

Selecting Winning Dogs with BrainMaker Neural Networks

Mr. Derek Anderson (Lakewood, CO) has trained neural networks that assist him in picking winning dogs at the racetrack. He trained the neural networks with two months of race results found in the daily racing booklets. Once trained, he runs the current day's race information through seven neural networks. He adds up the dogs' "scores" from his neural networks and places them in predicted finish order. Whenever the first place dog is ahead by at least ten neural network points over the second place dog, he bets on the winner. He claims 94% accuracy with this method, but he can bet on only a third of the races.

Mr. Anderson input information for approximately 300 races for the training file. The neural network looks at the statistics for three dogs at a time and outputs which of the three dogs did best. If there are eight dogs in a race, he must group the dogs in all possible combinations of three: dogs A, B and C; dogs A, B and D; dogs A, B and E; etc. For each race, there are 56 combinations, or sets of input data.

The data Mr. Anderson uses include the winning time of the race, the time that each dog took to finish the race, the time that dog reached each of four positions in the race (out of box, first corner, backstretch, outside corner) as well as comments about the dog's behavior. The behavior was classified as one of fifteen types such as ran wide, bumped, hit, and ran inside. He presented these pieces of information for each dog for each of the last eight races the dog ran. His networks have 504 inputs (21 statistics * 3 dogs * 8 races = 504).

Mr. Anderson has designed six basic neural network with these 504 inputs. The difference in the six networks is the output. One neural network has three outputs which represent which dog is best: dog A, dog B or dog C. The dog which was best gets 1, the others get a 0. Another has three outputs which represent which dog did worst of the three. The dog which did the worst gets a 0, the others get a 1. Another network has three outputs which represent whether the dog was in the top three finishing positions. These dogs get a 1, the others get a 0. An opposite network outputs if the dog was not in the top three. Another pair of networks output whether the dog was in the last three to finish the race.

Because the dog racing information is not available in computer format, Derek spent a lot of time doing data entry. When it's time to predict a race, Derek runs the data through all of his networks and adds up the score for each dog. The scores range from 0 to 25 most of the time. The dog with the highest score is the winner.

INPUTS
1st race, dog A: win time
1st race, dog A: finish time
1st race, dog A: 1st corner time
1st race, dog A: backstretch time
1st race, dog A: outside corner
1st race, dog A: behavior
1st race, dog B: win time
(etc for 8 races, dogs A B and C)

OUTPUTS
dog A rating
dog B rating
dog C rating