Bird Apprehension Forecasting with XGBoost

This article applied exploratory data analysis (EDA) and time seriesforecasting using extreme gradient boosting (XGBoost) to birdapprehension data from the Brazilian Institute of Environmentand Renewable Natural Resources (IBAMA) covering 2010 to 2024,the dataset processed included 150,000 records, filtered to focuson significant patterns across two distinct periods: a COVID-19pre-pandemic period (2010-2020) and a period that includes theCOVID-19 pandemic (2014-2024). This analysis identified shifts inapprehension patterns with an overall decrease of 69.17% in birdapprehensions during the pandemic. The XGBoost model demonstratedsatisfactory root mean squared error (RMSE) on most ofthe species except on Zenaida auriculata and Zenaida auriculatanoronha where the RMSE values were 37.73 and 47.80, respectively,for the (2010–2020) period, while for the (2014–2024) period, theRMSE values were 42.93 and 48.76, which are higher values whencompared to other results. The most apprehended bird species prepandemicwas Sicalis flaveola with 163, 437 apprehensions, whileZenaida auriculata and Zenaida auriculata noronha remained highlyapprehended throughout both periods. The northeast regions ofBrazil, with states like Rio Grande do Norte, Ceará and Paraíba,showed the highest apprehension rates. Overall, the model achievedsatisfactory performance in understanding the pattern of apprehensionnumbers in the tested periods, with significantly low RMSEvalues for the ten most apprehended species.