<![CDATA[Satellite imagery is utilized in various fields, one of which is land use and land cover (LULC) analysis. This study aims to classify water bodies using machine learning models such as SVM, K-NN, RF, CART, and GNB. The data source is obtained from the Google Earth Engine (GEE) platform using Sentinel-2 Level-2A satellite imagery, with a dataset of 5,514 data points per year. The Pixel-Based approach is used as the main method for data extraction, while CRISP-DM is applied as a structured methodology for data management. The parameter indices used include the BSI, NDBI, MNDWI, NDVI and AWEIsh. The results of these calculations serve as dataset features for training algorithms in the model development and training process. Each model has its own parameters, making parameter selection crucial in the training process. Model evaluation is conducted using a confusion matrix. Based on confusion matrix analysis, accuracy, precision, recall, and F1-score are calculated. Among the five models, SVM achieves the highest accuracy at 87%, followed by RF and K-NN. This indicates that the SVM model performs better in binary classification. Ground truth analysis is also conducted using the QGIS platform, which visualizes the classification results, with SVM providing the best visualization.]]>
Copyrights © 2025
