By Florian Marty
When we think of industries with a wide adoption of machine learning (ML), we first think of sectors such as healthcare, financial services, retail, and automotive industries. Farming and agriculture are usually not among the industries which come to mind.
However, there are many challenges which modern farmers have to overcome to increase their productivity while lowering the environmental footprint. Under these circumstances, it is not surprising to see that more and more farmers and countries are pushing for innovative solutions in farming and agriculture.
What (the hack) are hackdays?
“Hackdays” have a long tradition in the IT sector and industries which have led the data driven revolution. Generally, the goal of a hackday is to bring together a diverse range of participants with different expertise to solve a different pressing business challenges with new and innovative ideas. One could call it “organized thinking outside the box”.
Usually, the organizers present a set of challenges at the start. Then people organize themselves into interdisciplinary teams based on interest and expertise, and they start working on the challenges. At the end of the hackdays, the participants present a prototype solution to the challenge which they have they built during the hackday.
Open Farming Hackdays
The Swiss Opendata Association teamed up with the Landwirtschaftliches Zentrum Liebegg (an argicultural research institute in Aargau Switzerland) to organize the first ever hackdays to meet the need for innovation in farming and agriculture. Ahead of the hackdays weekend in early September, 18 different challenges were submitted to the organizers.
The challenges ranged from marketing and sales of wine, stopping land erosion, to using machine learning for decision support for artificial insemination (AI) of milk cows. Further below in the article we have a look at the individual challenges and explain where Modulos believes ML could add significant value.
Why Modulos AutoML is made for hackdays
Our vision at Modulos is to enable everyone with a sound understanding of the value of data to use ML to solve business critical problems… even if you are not a machine learning engineer or data scientist. The Modulos AutoML platform automatically selects the most applicable ML model, performs model architecture search, trains the model, and systematically tunes the hyperparameters resulting in a fast generation of multiple models for testing and deployment.
That makes it an ideal platform for hackdays with domain-level experts trying to solve their challenges by implementing machine learning. Furthermore, the platform is easy to use with little to no code work needed. That’s why it was a natural choice for us to support the Open Farming Hackdays by providing our AutoML platform to the participants.
Machine Learning suitable challenges at the Open Farming Hackdays
As mentioned, in total 18 challenges were submitted ahead of the Open Farming Hackdays. Each challenge was showcased by the organizer in a 2 minute presentation, followed by an in depth discussion at a poster. This allowed the participants to select the most interesting challenge and form the groups accordingly.
Before the Open Farming Hackdays, we were already in contact with the team from the Swissherdbook to prepare the data for the challenges Cow Value and Decision Support Besamung (engl. insemination). In short, the Cow Value challenge aims at giving milk cow farmers an estimate about the future economical value of a cow. This allows the farmer to make more economical decisions during a cow’s lifetime.
Tied to the Cow Value challenge is the Decision Support Besamung challenge. Quick primer on dairy farming: Simplified, a cow produces milk for about a year (305 days) after giving birth. As such the farmer has to decide each year if the cow should be subjected to artificial insemination to trigger a new lactation period. As artificial insemination has high costs for the farmers and is very stressful for the cows, a decision support system reducing unsuccessful artificial inseminations is desirable.
From a machine learning perspective the Cow Value problem can be seen as a form of regression problem while the artificial insemination problem can be seen as a classification (yes vs no) problem. Both types are supported with the AutoML platform including multiple individual models, respectively. Both these challenges relied on tabular data provided by the Swissherdbook.
Additionally, we also supported a challenge on the Früherkennung Milchkuh (early detection of disease milk cow) with a third team. Here again the data were provided by the swissherdbook.
Further, we thought of helping with the Smarte Bewässerung (Decision support watering) and Erosionsvermeidung (Prevention of erosion) as an ML-based system would be applicable. Both challenges could use the capabilities of the Modulos AutoML platform using image data, tabular data, and a combination of both data types as input data. Due to time constraints on site we did not focus on these challenges.
Takeaways from the Open Farming Hackdays (besides delicious apples)
There is a strong interest among Swiss farmers and more specifically the farmers from Aargau to search for innovative solutions to some of the biggest challenges they are facing. With over 70 participants 11 out of the 18 challenges were tackled during the event. Showing the true spirit of collaboration some of the marketing related individual challenges formed larger groups to come up with more impactful potential solutions.
In all involved challenges, we were able to generate prototype models using the Modulos AutoML platform. Clearly, it showed the importance of data quality, data preparation and the need for balanced data sets to generate meaningful models. Nevertheless, the presented solutions serve as a very good starting point for future work with our partners in the agriculture and farming industry. We are positive that many of the challenges tackled and the solutions presented at the Open Farming Hackdays will help Switzerland’s farmers in the future to achieve the goals of more productive farming while lowering the environmental footprint.