In November 2015, Syngenta released the call for Crop Challenge. It had a huge dataset at the disposal, which included weather, yield and soil characteristics at its farms in the US. The company decided to share the data with the community and get the answer to the question of seed selection. Competitors needed to develop an algorithm for choosing 5 out of 180 soybean varieties that should be planted on the Evaluation Farm to maximise yield and minimise risk.
One of the teams that recognised the significance of this opportunity was BioSense. This institute from Novi Sad, Serbia is concerned primarily with the use of IT in agriculture, which includes satellite image processing, development of agricultural sensors, communications and data analytics. Crop Challenge definitely falls within the scope of their research generally directed towards increasing food production with minimal inputs. They saw Crop Challenge as an ideal opportunity to test their technology in the field of seed selection and broaden their knowledge in this area. Their solution was based on a novel method called weighted histograms regression. It is a method based on voting. Yield at the Evaluation farm was predicted for each seed variety based on votes of the farms from the training dataset. However, not all votes were equally important. Each of them was weighted according to the similarity of the Evaluation Farm and the relevant farm from the dataset, so that similar farms would have a greater influence on the prediction than those with completely different soil characteristics. Now that the yields of all the 180 varieties were predicted, the optimal combination needed to be found. In order to do this, BioSense employed Portfolio Optmisation Theory. This is originally a theory from economics that deals with diversification of investments. In this use-case, rather than finding the optimal portfolio of assets in which money should be invested, it served for choosing the optimal portfolio of soybeans that should be planted on the Evaluation Farm to maximise the yield and minimise the risk.
Out of 500 applications and 30 teams that successfully submitted the solution, BioSense managed to get into the finals along with 5 other teams from MIT, Stanford and other universities and companies. The finals were held in Orlando, USA on 11th of April, when the competitors gave presentations about their solutions. They approached the problem from different angles, but what is common for all of them is the excellence of science involved. In this tough competition BioSense managed to win the fourth place, while the first prize went to Stanford University. This is an exceptional result for the Serbian institute. They say that it has been a wonderful experience to work on this problem and meet and exchange ideas with colleagues from around the world. They were truly delighted for having a chance to address one of the humanity’s biggest challenges – the problem of food security, by developing an advanced, innovative and environmentally friendly solution.