Deep learning models trained on more than 1000 penalty kicks in football matches can predict which way the ball will go better than real-life goalkeepers.

“Penalty kicks are some of the most decisive moments in soccer, often determining the outcome of major tournaments,” says David Freire-Obregón at the University of Las Palmas de Gran Canaria, Spain. “Despite this, real-time support for goalkeepers is still largely intuition-based. We wanted to explore whether machine learning could predict shot direction from a kicker’s body motion.”

So Freire-Obregón and his colleagues scraped 1010 penalty kicks from real, televised matches in Spain. Of those clips, 640 were deemed analysable by AI models, while the remainder were thrown away for being blurred, too short or obstructed.

Each clip was then fed into 22 deep learning models, which had to guess whether the penalty would go left, right or down the middle, based on the video footage and the simple fact of whether the player was right- or left-footed.

The best-performing model was able to correctly identify whether the ball went right, left or down the middle 52 per cent of the time – better than the 46 per cent accuracy real goalkeepers had in the matches. When the researchers removed the less-used middle option, model accuracy increased to 64 per cent – nearly 10 percentage points higher than human goalkeepers given the same information.

The researchers were surprised “by how much subtle motion cues, before the ball is even kicked, can reveal intent”, Freire-Obregón says. He hopes the information can be useful for goalkeepers in training, but using AI prediction in a match situation will be more of a challenge.

“What we aim to explore next is whether these probabilities can be anticipated in advance for penalty kicks, using only the kicker’s motion before the shot,” he says. “And if so, how early can such predictions be made while still maintaining meaningful accuracy.”