Blog post: Sugar content estimates using production data from a sugarcane mill

Data mining and artificial intelligence techniques have been shown appropriate to various prediction challenges: sports matches’ results, health problems, product demand in supermarkets, lip reading. In this work, we show that crop’s responses to the environmental factors that interfere in its growth may be estimated by data mining techniques. The particular crop we studied is sugarcane and its response, sugar content in its culms.

We used data mining techniques and the database from a sugarcane mill and showed how these databases have been underutilized: not only techniques are able to identify variables that are often recognized as those that most influence in sucrose storage, but also errors on estimates are very low.

The advantage of the use of these techniques is that when one uses a database that comprises several years, so that the crops were subjected to diverse weather conditions, estimate errors will be lower, even for atypical years, due to the techniques' ability of responding to diverse patterns. With a database comprising two years, we reached errors as low as 5.4 kg of TRS per ton of sugarcane in 90% of our predictions.

If this methodology is to be used by mill managers, they may take advantage of good production estimates without additional information, depending, obviously, on the quality of weather forecasts available. This could lead, for example, to better prioritization of harvesting areas.

If you want to know more of our work, you can find it in here.