A Meta-Feature Model for Exploiting Different Regressors to Estimate Sugarcane Crop Yield

The crop yield prediction is crucial for the sugarcane grower to estimate the amount of biomass that will be harvested in decision-making for the acquisition of agricultural fertilizers and pesticides, for carrying out the harvest, and for the reform of the cane field. Usually, the features used for crop yield prediction are based on the direct observations of what occurs on the field collected by sensors or manually. But modeling the problem with new features, calculated by regressions applied to features collected from the phenomenon, can help to explore better the results that dataset retrieves. And it is possible by using these retrieves as new features to be modeled in other regressions. This article explores the viability of producing new features, called here meta-features (MF), to find better results for the sugarcane crop yield prediction. These meta-features were created from the results obtained by different regressors used to analyze which of them would present the best prediction in the original dataset. The regressions using these meta-features obtained better results in terms of ${\bar R^2}$ and errors associated with the crop yield measured on the field.