Using data to improve sports performances

‘Tell me something I don’t know,’ said skating coach Jac Orie to Leiden data scientist Arno Knobbe. And he did. Knobbe and his colleagues now assist athletes in all kinds of ways with the help of data mining.

It all started – in his own words – as a ‘sabbatical project’. As a specialist, Arno Knobbe had always had a keen interest in applying data mining techniques in practical situations and in developing those techniques based on real-life problems. For example, he had already worked on projects to monitor motorway bridges and to comb through medieval documents. He was also fascinated by data on the performance of the human body: how does it work, and how does it react to stimuli?

Skating professor

Around six years ago, he met coach and ‘skating professor’ Jac Orie (three-time winner of the Sports Coach of the Year award), and explained how data mining could extract surprising insights from large amounts of data – in this particular case, data about skaters on the Jumbo-Visma team. ‘Tell me something I don’t know,’ challenged Orie dryly. ‘It was the perfect project,’ laughs Knobbe

He set to work on the training data and race results, which the team had recorded in meticulous detail. He then used that data to carry out a personalised analysis of each of the skaters and make recommendations for their training programme. ‘For example, for Kjeld Nuis I discovered that, in the five days leading up to a competition, it was better for him to limit his morning training sessions on the ice. Otherwise his muscles were still too tired on the day of the race, which had a negative effect on his performance.’

Following this initial study, the collaboration between the Jumbo-Visma skating team and Knobbe’s research group intensified. The team now participates in, and sometimes co-finances, new projects every year. And, of course, the research results are also very valuable for the skaters. Orie and his colleague Nico Hofman also often co-author scientific articles.

Speed in the bend

Knobbe’s research group is the only one in the Netherlands that conducts this kind of comprehensive research into sports and performance using data mining. The group offers its analytical and sports services through the Sport Data Center, which combines the expertise of data scientists and movement scientists. Knobbe: ‘We used to just analyse data that was provided to us by the sport, but recently we have also started to collect data ourselves. For example, we use the measuring skate, a skate packed with sensor technology that you can use to collect a huge amount of data about the skating stroke. The latest generation of measuring skate was developed by Leiden, VU Amsterdam and Viking.’

‘In another project, we are developing a new method to determine the speed and line of skaters in the bend. You can’t do that with measuring loops in the ice, which is what we have at the moment. The way in which a skater approaches a bend influences the moment at which they pass the loop, which makes it difficult to get an accurate measurement. To get a better idea of the line, and to pinpoint exactly where a skater slows down and accelerates, you need to be able to accurately locate the body and individual skates through the bend. We want to start measuring skaters’ speed using a new suspended camera tracking system. This is the best angle for following skaters through the bend without arms or legs or other skaters getting in the way. The plan is to install five professional cameras in the ceiling at the Thialf skating arena and use computer vision to digitalise skaters’ movements.’

Come a cropper

Although skating is the main focus, Knobbe’s group also works on other sports. For example, they help the Dutch volleyball team to analyse training data and report injuries in order to formulate personalised advice on how to prevent injuries. They help a well-known Dutch tennis player with tactical analyses, collecting and analysing data from every rally. The Jumbo-Visma cycling team also collaborates with Knobbe. Last year, for a well-known cyclist Knobbe analysed which altitude stages of the Tour de France were most and least suited to him, based on three years of power, GPS and heart rate measurements. This year, the team analysed the cyclist’s preparation for key races. ‘For example, we looked at the wattage that the cyclist produced. We came to the conclusion that if this cyclist spent too much time above a specific threshold of power on average over the four days before the race, he would probably come a cropper on the fifth day.’

Knobbe is reluctant to say whether his analyses are the decisive factor for better sports performances. ‘Let’s just say that top-level athletes are constantly trying to improve their performance, one percentage point at a time. And we contribute towards that.’