CAN YOU TRUST THE METRICS? DATA RELIABILITY IN MACHINE LEARNING FOR SÃO FÉLIX DO XINGU

Mapping the Amazon Rainforest presents significant challenges due to its dense vegetation and the difficulty of accessing various regions. We used the Balanced Random Forest (BRF) algorithm to classify an area near the Carajás Province and generate a lithology map based on the previous mapping by the Geological Survey of Brazil (SGB). The BRF algorithm has proven itself as reasonably reliable on identifying different lithology classes and could represent a powerful tool towards quality assurance of already mapped areas or for the improvement of preliminary mappings already conducted on areas with low data. On the other hand, it evidences the ever-growing need for a widespread guideline towards data collection and handling within the geosciences.