ABSTRACT
Forest fires burn large areas of native vegetation and it causes impacts in the social, economic and ecological scope. Burnt areas classification can help understand fires occurrence and support public policies. This work aims to develop a method of automatic burnt areas classification. The method is based on the application of Automated Machine Learning in data sets, from the Landsat-8/OLI satellite images of 2018 and 2019. We intend to answer the following research question: “Is it possible to automate the choice of machine learning models and maintain quality levels in the classification of burnt areas?”. The contribution of this research is to determine whether a predictive model, trained with validated samples from 2018, is capable of classifying fires occurrences in 2019. For the performance evaluation, the following metrics were analyzed: precision, probability of detection and average success rate. The results indicate that the method has a high potential to classify burnt areas.