On the eve of the start of the Rugby World Cup, there have been whispers of teams spying on each other. Unavoidable gamesmanship, perhaps, but there’s no doubt that cheating in sports is a problem that officials struggle to deal with.
Our new machine learning model could be a game changer when it comes to detecting questionable behavior and unusual results – particularly the practice of match fixing.
At present, the act of altering the outcome of matches for personal or team advantage is largely carried out by distortions in the sports betting market. When bookmakers notice unusual odds or changes in betting lines, they alert regulators.
But this approach is limited and often fails to identify all match-fixing, particularly in less popular sports or leagues. This is where machine learning can help.
Essentially a subset of artificial intelligence (AI), machine learning acts as a digital probe: mining sports data, uncovering hidden patterns and flagging unusual events. Machines can detect team performance and unexpected fluctuations by exploring all aspects of a sporting event.
Using AI to detect unusual activity
As part of our research, we introduced the concept of “anomaly match detection”, which involves identifying irregular results in games, regardless of the underlying causes.
The reasons can range from strategic losses to future profits – such as the practice of “tanking” in the US National Basketball League (NBA) – to marketing strategies to boost ticket sales, or a single day of poor performance.
Our research model allows us to flag unusual game results and refer them to regulators for deeper investigation. By leveraging machine learning, we can detect unusual matches by comparing our predictions to actual game results.
Read more: Why police should use machine learning – but very carefully
When we discuss sports anomalies, we are talking about matches that differ from the norm.
Match fixing – the deliberate manipulation of results for advantage – is one possible explanation for unusual game results, but not the only one. Recognizing the multiple reasons behind unusual match results can help improve our understanding of the complexities of the game.
In the wake of an unusual or unexpected result, spectators and officials may ask themselves: is this the result of an unexpected strategy or are there other influences on the game?
Christian Peterson/Getty Images
Learning from basketball
Our research methodology involves training machine learning algorithms to detect patterns in specific past events and subsequent game outcomes.
Once these relationships are established, algorithms can predict possible future outcomes. Discrepancies between these estimates and actual results could potentially represent abnormal matches.
To test our model, we looked at whether there are no common matchups in the 2022 NBA playoffs. We built a model to predict match results using data from 2004 to 2020 and then compared what the machine predicted with actual game results.
We found several inconsistencies in the 2022 playoffs, especially in the series of games between the Phoenix Suns and the Dallas Mavericks. In seven games against each other in May 2022, Dallas won four games and Phoenix won three.
Read more: Who will win the 2023 Rugby World Cup? This algorithm uses 10,000 simulations to rank the competitors
According to the data, the odds in the 2022 playoffs include a 0.0000064 probability of the Suns and Mavericks in the semi-final series of the NBA’s Western Conference – which consists of 15 teams.
We identified several players who performed during the playoffs that were particularly unusual based on data from their previous games.
It cannot be said that there was match fixing. Instead, our results flagged games and players who could then be pursued by regulators So Match fixing was a concern – which it wasn’t, it was just an example to test the model.
This approach to finding anomalies in a series of matches can be applied to many sports.
Scrutinizing a large number of anomalies can provide valuable insight into unusual match occurrences, helping regulatory bodies and sports organizations conduct thorough investigations and maintain fair competition.
Building faith in sports
Although our study focuses on specific sports, the principles and techniques may extend to other areas.
Studies show that machine learning can be used to protect the integrity of sports events and help regulatory bodies, sports organizations and law enforcement agencies maintain fairness and public trust.
But as we embrace the potential of machine learning, we must also navigate the ethical implications and ensure its transparent use.
In the future of sports, artificial intelligence can become an ally of fans, helping to ensure a level playing field where talent excels and spectators enjoy the authenticity of sporting events.