How AI Tries (and Fails) at Predicting Sports Games
Photo Source: PxFuel
Sports fans have no rival when it comes to the passion and devotion they can show for their teams and players. Be they devoted baseball fans, diehard hockey fanatics, or rabid soccer supporters, everyone wants to know if their team has a chance of winning it all—to be champions. Many fans will take pride in their city or country when their teams rise to the top and win it all. Additionally, many fans also bet on sports games, compounding their interest in the outcome of a game. Part of the thrill of watching sports is not knowing who is going to win, and while some have tried, artificial intelligence has not yet perfected the art of predicting sports outcomes.
With so much fervor over sports, sports journalists and oddsmakers spare nothing when talking about sports and making predictions as to who will win and why. However, even the experts with so much talk and rumination cannot consistently predict who will win their league or tournament. Professional handicappers who predict the outcomes of games and contests seldom do much better than 50% accuracy. According to betfirm.com, their all-time best handicapper, Jack Jones, has a winning percentage of 52.9% after over 10,000 games. According to an article titled "Meet the world's top NBA gambler," Bob Voulgaris, who made millions of dollars betting on basketball in the early 2000s, had a winning percentage of 70% until he lost his edge and a great deal of money. (espn.com)
Handicappers, predictors, and betters look at data to help make decisions, and sports are awash in data. Everything gets measured from player statistics such as speed, height, weight, time on the field, number of goals, amount of touches of the ball to team stats such as home and away wins and losses, historical performance against different teams, and on and on… Given that sports have rules and tons of data. So it makes sense that artificial intelligence, which gobbles up data to make predictions, would predict sports outcomes better and faster than any expert. Of course, people have limits to the quantity of data they can consume and understand, but computers have a much greater capacity.
In "Predicting Sports Results with Artificial Intelligence – A Proposal Framework for Soccer Games," published in Procedia Computer Science, the authors note the application of different types of artificial intelligence to sports predictions with varying results. (sciencedirect.com) For example, the Open International Soccer Database creators held a competition to see who could create the best AI prediction model using their data from over 216,000 soccer games. The best-performing program in the contest had a success rate of only 51.9%. Subsequent models that added more information, such as individual player and manager statistics, achieved higher accuracy. For example, Runzuo Yang's model for English Premier League soccer had an 81.8% accuracy rate for the 2018/2019 season, but the model could not extend to the next season.
More recently, the Swiss sports intelligence company—Sportradar—used their sophisticated AI system that simulates real play to predict the outcome of the Euro2020 soccer tournament. Euro2020, which had been postponed a year because of the pandemic, occurs every four years and crowns the best national soccer team in Europe. Using their simulated reality system, Sportradar writes, "We've tapped into the breadth of our technical capabilities to simulate the tournament, processing millions of data points from the last 20 years in order to identify the winning team." Before the start of Euro2020 in June 2021, they simulated the complete tournament 40,000 times. They concluded that the Czech Republic would beat Denmark in the final and that England would be eliminated in the semifinals. (inews.co.uk) The tournament ended July 11th with Italy defeating England in the final and the Czech Republic getting knocked out in the quarterfinals by Denmark!
Sports fans have no rival when it comes to the passion and loyalty for their teams and players, whether at the local level of a college team or professional franchise to the national level. However, part of the excitement of a game or the beginning of a tournament rests in the unknown—who will win it? Fans and those who bet on games all want to know the answer to that question, and handicappers have developed many systems to predict the outcome of games and contests. Considering that some of the best oddsmakers struggle to do better than 50%, it makes perfect sense to apply the power of artificial intelligence to predict sports outcomes. A survey of AI techniques shows some successes, but nothing yet comes close to perfection. Even with a massive amount of data and sophisticated simulation for the Euro2020, Sportradar completely missed the outcome. So, even with AI and a growing sea of sports data, fans still need to watch their teams to see who wins.
Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.