Automatic Identification of Spurious Records in the Rural Environmental Registry via Machine Learning

Abstract

The Rural Environmental Registry (CAR) is a mandatory electronic registry for all rural properties in Brazil. Currently, the CAR is gaining importance in land use monitoring and environmental regularization. However, the regularization analyses of the registry are done manually, requiring a significant amount of time and skilled labor for this task. This study aims to use machine learning algorithms for the classification and identification of erroneously filled (spurious) records. As a database, real CAR records previously manually labeled were used. Subsequently, six classification models were applied, and the generated results were evaluated and compared with each other. The classification models, in general, achieved evaluation metrics above 80%, and the Gradient Boosting obtained all metrics above 90% for the test dataset. The results obtained through the models demonstrated the feasibility of implementing the model in a real CAR scenario, through a REST API. Thus, creating the machine learning module in an API, it is possible to develop an application to agilize and improve the processes in the Rural Environment Registry in Brazil.