Comparative analysis of Random Forest and Support Vector Machine for LULC classification in Tashkent Region using Landsat-8 imagery
Annotatsiya
Accurate Land Use/Land Cover (LULC) classification is essential for effective environmental monitoring, sustainable agricultural management, and informed urban planning. With increasing land transformation driven by urbanization, deforestation, and climate variability, reliable classification methods are needed to support data-driven decision-making. This study presents a comparative analysis of two widely used machine learning algorithms—Support Vector Machine (SVM) and Random Forest (RF)—for LULC classification in the Tashkent Region, Uzbekistan, using Landsat-8 Operational Land Imager (OLI) data for the June–July of 2024. The workflow involved critical preprocessing steps, cloud filtering, and band selection, followed by classification using six dominant LULC classes: water, bare land, built-up areas, cropland, snow, and forest. Ground truth samples were used to train and validate the models. Accuracy assessment was conducted using a confusion matrix, and performance was evaluated based on Overall Accuracy (OA) and the Kappa Coefficient (KC). The results revealed that the SVM classifier slightly outperformed RF, achieving an OA of 96.78 % and a KC of 0.96, compared to RF’s OA of 94.95 % and KC of 0.94. The superior performance of SVM is likely due to its effectiveness in handling high-dimensional feature spaces and separating non-linear class boundaries, especially in heterogeneous landscapes like Tashkent Region. While both algorithms showed strong potential, SVM demonstrated better precision in classifying cropland and forest areas. These findings highlight the importance of algorithm selection in remote sensing-based LULC studies. The study contributes to ongoing efforts to enhance land cover mapping accuracy using machine learning and offers valuable insights for land managers, urban developers, and environmental policymakers. Future research may consider the integration of multi-seasonal imagery, ancillary environmental data, and deep learning frameworks to further improve classification performance.