How AgroCares uses machine learning to make recommendations more accurate

Sam Sarjant works as an ICT Software Developer at AgroCares. In his job as a developer he works with cutting edge technologies involving machine learning on a daily basis. In this blog Sam explains the potential of these technologies for tomorrow’s agriculture. But what is machine learning? How does it work? And why is it so important?

Machine learning makes AgroCares’ recommendations more accurate
Agriculture today faces many challenges. The increasing population and its concentration in urban areas drives up the demand for food. At the same time, environmental issues such as climate change require adaptation and resilience in farming systems. Resources such as mineral fertilizers can be wasted because of inaccurate recommendations. They are too often designed at large scale without taking soils complexity into account. AgroCares strives to make recommendations more accurate and calls upon machine learning to do so.

What is machine learning exactly?
Machine learning is a field of computer science that uses models and algorithms that learn from and make prediction on data. In other words, it produces predictive analyses thanks to models and algorithms that have the ability to “learn”, that is that their performance can improve without being explicitly programmed.

What does it have to do with agriculture?
Most farmers in the world receive recommendations on inputs application for their fields. Most commonly, they are based on wet chemistry analyses that are accurate but also costly and time-consuming. In developing countries, they are often made at regional – if not national scale. The role of predictive science and machine learning in agriculture is to link a database of wet chemistry analyses to quick spectrometer tests from the field. The results are fast, accurate and affordable recommendations.

How does it work?
Let’s take the case of soil analysis. To measure the composition of a soil, we first need to calibrate the country or region we work in. It consists of wet chemistry analyses on representative soils samples of the area. This gives us a database. In the meantime, we analyze identical samples with our own spectrometer tools (Lab-in-a-Box and Scanner) and that creates a second database. Thanks to empirical models, we are able to link both database and produce powerful prediction models. Therefore, quick spectrometer tests combined with our prediction models return accurate soil measurements.

The role of machine learning in our prediction models
Our prediction models get more accurate every time we add a sample to our database, and this is machine learning. In order to predict the content of an element in a soil sample, the model reaches into our database of all our previous sample tests, chooses the most similar experiences, combines them, and uses the combination to make a prediction. Because the model evolves with the database, and because the database becomes richer, the accuracy of the results improves constantly.

Is machine learning accessible for farmers?
Yes, it is. By combining easy-to-read analyses and recommendations with handy tools, AgroCares products put the power of predictive sciences in the hand of every individual.

So why is machine learning the future of agriculture?
New technologies involving machine learning benefit from the high power capacity of computers, which allows quick delivery of great and improving accuracy results. The method we use at AgroCares is adapted to soil, feed and leaf analyses, and is already approved by many farmers around the world. The recommendations generate better yields with more appropriate inputs … Can this not be the future of agriculture?