On a winter day four years ago, Takeshi Watanabe, then a graduate student at the University of Tokyo, went to watch the year-end Arima Kinen horse race at Nakayama racecourse in Funabashi, Chiba Prefecture. For him the day was more about watching horses and experiencing betting, rather than making money by gambling.
Six months later, however, he found he had become a fan of horse racing and enjoyed betting. It did not take long before he came up with an idea to mix his new hobby with his major at school, machine learning and analysis. That was how he started to develop an artificial intelligence (AI) horse racing prediction system.
“I had no knowledge of horse racing or betting, so I was not a good bettor in the beginning,’’ said Watanabe, CEO of AI and web service company RHT LCC that runs the prediction site AI Ringo. “One of my majors was analysis, and I wanted to see how my studies work in the real world.’’
AI prediction is all about analyzing data. Watanabe tries to get as much information as possible from various sources.
“We feed a variety of data, such as race results for the past 10 years, weather, racecourse conditions and more, into our AI. By analyzing the data, we predict which horse has the best chance to win,’’ Watanabe explained. “The process of prediction is almost same as what people do. But the biggest difference is that AI can analyze massive amounts of data and find the tendencies.’’
To better understand the system, we can look at the 11th race of the Muromachi Stakes that took place on Oct. 24 at Kyoto Racecourse (1,200 meters, dirt, three-year-olds or older) as an example.
AI Ringo recommended making a quinella place bet, where a bettor wins the bet when two horses chosen finish in the top three in any order. AI Ringo predicted the pair of horses numbered 2 and 6 or 2 and 8 had the best chance to win the bet.
The Muromachi Stakes was won by Red le Zele (horse number 2), followed by Ryuno Yukina (9) and Lord Lazurite (6). The AI Ringo prediction of the 2 and 6 horses was right and the payout was ¥1,420, meaning if a person bet ¥100 on the 2 and 6 horses, they would receive ¥1,420 for the bet.
If someone bet ¥100 for each of the two pairs of 2-6 and 2-8 for the race, they would be ¥1,220 richer than before the race. In this case, the return ratio is 710% (1,420 ÷ 200 x 100).
Watanabe said AI Ringo’s focus is not only predicting which horse to win, but how much a bettor gets in return.
“If you want to just win a bet, you can bet the top favorites. You could win 80 to 90% of them, but the return ratio of those bets is low. If the payouts are ¥110 for ¥100, your profits is small,’’ Watanabe said. “But even when the winning percentage is about 50% and payouts are 3-to-1, you have a chance for big returns. We put more value on return ratio.
“Local races tend to have higher return ratio than major races because a lot of information is provided for major races. That makes prediction easier,’’ Watanabe said. “But for local races where the AI prediction is not common yet, there is a chance to see good returns.’’
Because the AI prediction is based on data, selecting the important data is crucial, Watanabe said.
“The data for an irregular race, such as those where a horse doesn’t finish, could cause a bug in the system. We can’t put all the data we have. We have to “clean” the data first,’’ Watanabe said. “Cleaning means abandoning unnecessary data and leaving the important data. We need horse racing knowledge to do this effectively and this is the most difficult thing in about AI prediction.’’
Watanabe expects the AI predictions will become more popular and more accurate. Unfortunately, more accurate predictions lowers the odds and causes returns to go down, leading to AI Ringo potentially losing customers.
In a time of both misinformation and too much information, quality journalism is more crucial than ever.
By subscribing, you can help us get the story right.