Comparative Analysis of Machine Learning Algorithms for Land Cover Classification Using Sentinel Imagery

To address the insufficient accuracy of land cover classification in tropical agricultural regions of Southeast Asia, this study systematically compared three machine learning algorithms - Support Vector Machine (SVM), Classification and Regression Tree (CART), and Random Forest (RF) - on the Google Earth Engine platform for their performance in complex land cover scenarios. The results demonstrate that the Random Forest algorithm achieved optimal performance in nationwide land cover classification for Vietnam, attaining an overall accuracy of 92.36% (Kappa coefficient = 0.851). This represents improvements of 7.94% and 4.19% over SVM (OA = 85.42%) and CART (OA = 88.17%), respectively. Quantitative evaluation based on confusion matrices revealed that our method significantly enhanced the identification capability for cultivated land categories, achieving a producer accuracy of 91.23%, user accuracy of 84.69%, and F1-score of 87.87%. Compared with the European Space Agency’s WorldCover product, the classification accuracy for cultivated land improved by 31.6%. This study provides a reliable algorithm framework for land cover classification in tropical agricultural landscapes.