Premier League picks: Machine learning for fantasy football
Millions of players in the UK compete in fantasy football leagues, tinkering with their squads to generate maximum points from their teams. Shrewd dealings in the transfer market can make or break team’s seasons and entrants scramble week on week to strengthen their sides as Premier Leagues stars go on and off the boil.
Dr Sarvapali (Gopal) Ramchurn, an Associate Professor in a world leading artificial intelligence (AI) research group at the University of Southampton, is a seasoned fantasy football competitor and has innovated a tech secret to getting a team into the top one per cent of the field.
“It’s so addictive,” he says, “picking my team, making transfers and competing with my workmates. There’s a real skill to making the right decisions and people would do anything to move their team up the league.”
Gopal has created cutting-edge algorithms for companies like BAE systems, been part of multi-million pound AI projects and written around 100 papers on AI and machine learning during his academic career. A few years ago, he and Electronics and Computer Science (ECS) postgraduate Tim Matthews decided to channel this expertise to train a computer to help optimise fantasy football team and transfer selection.
Fantasy managers can compete with a Squadguru AI-fuelled 11 in a ‘Challenge the Squadguru’ league in fantasy.premierleague.com by entering league code 2917382-677658.
The system was built using two steps. The first harnessed Bayesian Machine Learning techniques and five years of past football data to create and train a predictive model. The model was used to forecast the outcome of future matches and, more crucially, the performance of individual players. An advantage of using the Bayesian approach was that the academics could combine expert knowledge with player data to increase the accuracy of estimates.
They next created a combinatorial optimisation algorithm which worked out the best transfers to make given the allowed budget and other constraints on teams that can be formed. The algorithm worked out which transfers would maximize the return on an investment.
The tool was tested on player data in previous seasons and produced teams which would have consistently ranked in the top one per cent of the Fantasy Premier League. In this year’s competition of 3.3 million players worldwide, this would mean finishing in the top 30,000 teams.
The algorithm was then unleashed on live games and Gopal connected with Paul Morgan from the FantasyFootballFirst blog. He began to provide transfer advice to the subscribers and soon had over 30,000 people using the predictions every month on the most popular fantasy football advice blog in the country.
Squadguru is currently being used for free, but it has become clear that people would pay for the service so Gopal has decided to spin out the technology into a new startup.
A subscription-based service would offer sound advice to ensure players are well placed to make good decisions. Investors interested in exploring what AI can do for team and transfer selection are encouraged to get in touch with Gopal using the contact form on this page.
The system needs investment to hire staff to develop front end and back end systems.
There are millions of potential customers for the service following the Premier League. With over 40 million people playing fantasy sports in the US and Canada, there is also scope to expand Squadguru to tap into lucrative markets overseas.