Researchers from IDLab, the imec research group at the University of Antwerp, have developed a computer system that attempts to predict race results in daytime races based on historical performance data and artificial intelligence (AI). Forecasts can be tracked from now on wiewintdekoers.be. For the Flanders tour, the AI system predicts that Matthew van der Paul will be the first to cross the line.
Artificial intelligence changes every sector. An exponential explosion of data, combined with increased computing power and more efficient algorithms, makes it possible to make more informed decisions. This also applies to the world of sports. As cycling data (race scores, strength, heart rate, …) becomes more widely available and in a more structured way, more data-based predictions can be made. To demonstrate the potential of AI in sports, IDLab is now launching the wiewintdekoers.be website.
The results of the race are difficult to predict. They depend on the type of race, the condition of each racer, the strategy of their team, the weather conditions, as well as the unexpected course of the race. However, researchers from imec and UAntwerpen have successfully developed an AI algorithm that can preview the results of one-day races.
Based on data from the past twenty years, the system has taught itself what are the most important predictive parameters for predicting the outcomes of a particular classic spring (in the elite men on the road class). For example, the AI model automatically detects a performance that is important for a good performance in a race like the Flanders Tour. Based on the performance data of all cyclists over the past three years, the algorithm then tries to predict the complete ranking. Since everything is automated, this can be applied to any one-day course.
The first version of the system was evaluated based on all classic bikes in 2018 and 2019. In it, the AI scores slightly better than human cyclists’ expectations. For example, the computer model was able to correctly predict seven of the top ten passengers on the 2018 Flanders Tour, including winner Nikki Terpstra.
The AI is learning from the patterns it recognizes in similar bike racing sequences. “The pandemic and the changes to the 2020 cycling calendar were no gift in that regard. Our computer model was completely confused,” sighs Professor Stephen Latrey (imec / UAntwerpen). However, in the meantime, the cycling calendar is again more natural and the algorithm has been expanded with additional parameters. For Gent-Wevelgem, last Sunday’s race, the AI system correctly predicted four of the top five contestants, including winner Wout van Aert.
Researchers at IMEC and UAntwerpen want to improve the system even further in the coming years, for example by taking more injuries and falls into account. By summer, the scientists want to expand the system from single-day races to theater races, so they can also predict the Tour de France ranking. It should also be possible to predict how the flight will take place.
Using wiewintdekoers.be, IDLab demonstrates the potential of AI in cycling, but the possibilities for applying the technology are much broader. “Our methods can also be used to track young cycling talents more quickly, which is something we are already working towards with professional cycling teams. When they provide us with additional data, we can help them make better training and employment decisions. More and more data is also generated in many Other sectors, from which valuable insights can also be obtained. Artificial intelligence is ready to change any sector, ”Stephen Latrey concludes.