Want to know the score of the football game before it actually begins? We created the OneSoil Sports app for that.
Yes, you heard that right — we’re an agricultural startup, and we’ve learned how to predict football match results with 75% accuracy by the vegetation index of the playing ground. What started out as a joke, turned out to be a disruptive approach to football. We have analyzed 150 playing grounds and 12589 games. In other words, we trained the neural network to think as a professional coach. Now we’re ready to reveal the full story — involving AI, one South American football league and the new OneSoil Sports mobile app.
At OneSoil, we develop apps and a web platform for precision farming. To put it simply, we use satellite data and machine learning algorithms to measure various field indicators and provide farmers with precise analyses and forecasts via devices. First, we collect data, a lot of data, and then develop and train neural networks to analyze the information the way you need. For all of our web tools, we use satellite images provided by the Copernicus programme and ‘ground truth’ data from fellow farmers across Europe. In 2018, we allocated all the fields in Europe and the USA and created the OneSoil Map to visualize each of the 60 million of them.
While we were preparing the OneSoil Map, the head of our R&D department Alexander Kalinovsky noticed that not all the fields we had detected were agricultural fields. A big part of them, around 19%, were actually football fields. This year we are testing our models in the rest of the world, and in South America, this number is even higher, around 28%. Just for fun, our team decided to see what happens with the NDVI index on the playing grounds.
The NDVI (Normalized difference vegetation index) is an indicator of a plant’s health based on how the plant reflects different light waves. In precision farming the NDVI index is used for everything — it really is a ubiquitous tool. You can check plant development, the existence of diseases, density, detect areas of low and high yield, choose the best time for the application of seeds, fertilizers, and pesticides. Also, you can understand when it is time to harvest the crop. Coming back to the football fields, we decided to compare changes in the NDVI index of the field with the scores of the matches played there.
For the experiment, we selected 150 fields in Belarus, Ukraine, France, and England. We analyzed the change of the vegetation indices for the last 3 years and compared them with the results of 12589 games. First, we didn’t believe the results of the analysis, but it turned out to be a matter of fact:
the team that is playing on the part of the field with a higher vegetation index more often — wins, while the one that is playing on the part with a lower vegetation index — loses.
After we tested our hypothesis for past games, we decided to see if our model worked for future games also. Using machine learning algorithms, we have analyzed game tactics of all the European football clubs.
In other words, we trained the neural network to think as a professional coach.
In combination with the NDVI index analysis, that allows us to predict the results of any football game with 75% accuracy. Now, it works only for one year ahead but we plan to improve the algorithm. Unfortunately, we’re unable to reveal all the details of the analysis due to our obligations to the one football league from South America that became interested in the technology. In collaboration with them, we plan to release a new app for professional coaches at the end of the year. And for all the football fans, we will release a free OneSoil Sports mobile app this summer.
You may be saying to yourself right now, ok, you can predict the score of a football game using the NDVI index but what exactly happens on the field? It is a good question, and just like in precision farming, we don’t know for sure. You have to go directly to the field and find out the reason for such field conditions yourself. We have a couple of ideas. We assume that in the zones with a high NDVI index, the grass quality and density is higher, so there is better traction between the grass and the football boots of the players. This improves the speed and mobility of the player. Another hypothesis is that it is the relief of the land that makes all the difference. There are tiny changes in the relief of the playground that are not visible to the human eye but are recognized by AI. Currently, our specialists are finishing the development of an algorithm that allows building a relief map using the vegetation index (you can see the example in the picture). If you want to test our new technology, please email us at hello@onesoil.ai.
‘This is all quite unexpected for us. However, we are happy to help football fans, coaches and professional players all over the world. In addition, many people from our team are football fans so we really enjoyed working on this new product. Later on, we plan to develop similar tools for score predictions in golf, rugby, cricket, and field hockey’, says Slava Mazai, OneSoil’s CEO
P.S. Happy April Fools’ :)
