Comparison of random forest and extreme gradient boosting algorithms in land cover classification in Van Yen district, Yen Bai province, Vietnam

Abstract

Land cover classification using remote sensing data plays a crucial role in resource management and environmental monitoring. This study compares the performance of Random Forest (RF) and Extreme Gradient Boosting (XGBoost) algorithms in land cover classification in Van Yen District, Yen Bai Province, Vietnam. The input data include Sentinel-1 synthetic aperture radar imagery, Sentinel-2 optical imagery, and a total of 7,214 sample points used for model training and validation on the Google Colab platform. The results indicate that both RF and XGBoost achieve high classification performance, with overall accuracy ranging from 94.8% to 96.3% and Kappa coefficients between 0.936 and 0.955. Notably, RF demonstrates greater stability and consistently higher accuracy than XGBoost in both scenarios: using Sentinel-2 data alone and combining Sentinel-2 with Sentinel-1 data. The findings provide a scientific basis for selecting suitable algorithms and data sources to improve land cover classification efficiency in the study area.