Spatial Classification of Sentinel-2 Satellite Images with Machine Learning Approach

Urban expansion and land use change are increasingly critical issues in developing regions, where rapid development often leads to unplanned growth and environmental challenges. Accurate and timely classification of built-up and non-built-up areas is essential for supporting sustainable spatial planning and resource management. This study aims to classify built-up and non-built-up areas from Sentinel-2 satellite imagery using a machine learning approach and to analyze their spatial distribution around the Universitas Islam Indonesia (UII) campus. Three machine learning algorithms—Support Vector Machine (SVM), Logistic Regression (LR), and Decision Tree (DT)—were applied to perform the classification, and their performances were evaluated using four metrics: accuracy, sensitivity, specificity, and Area Under the Curve (AUC). Among these, the SVM method demonstrated the best performance based on the highest average accuracy, the smallest variance difference between training and testing datasets, and consistent results across multiple iterations. Using the classification results from the best-performing model, a spatial density proportion analysis was conducted. The findings revealed a clear spatial trend: areas closer to the UII campus exhibited a higher proportion of built-up land, while areas located farther from the center had a greater share of non-built-up land. These results confirm the effectiveness of the SVM algorithm for land cover classification using Sentinel-2 imagery and offer valuable insights into urban development patterns in the study area. The outcomes of this research can inform urban planners and policymakers in developing data-driven strategies for sustainable land use, infrastructure development, and campus-centered regional growth planning.