Assessing Supervised Classification Algorithms for Mangrove Detection and Differentiation from Other Forest Types in Paiton District, Probolinggo Regency

Remote sensing provides an effective approach for monitoring forest cover dynamics, yet distinguishing mangroves from other forest types remains challenging in heterogeneous coastal landscapes. This study compares six supervised classification algorithms: Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Mixture Discriminant Analysis (MDA), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) for mangrove detection in Paiton, Indonesia, using Landsat 8 imagery (2015–2025). Predictor variables included NDVI and MNDWI, with model performance evaluated via the Kappa coefficient and F1-score. Feature importance was analyzed using Mean Decrease Accuracy (RF) and Gain (XGBoost) to identify the most influential predictors. The results show that SVM achieved the highest average Kappa (0.93). RF and XGBoost demonstrated the most stable performance, with consistently high F1-scores and low standard deviations (0.05). In contrast, LDA and QDA showed lower and less stable accuracy. Overall, RF and XGBoost are recommended for multitemporal LULC classification, as they effectively utilize vegetation indices to separate mangroves from spectrally similar forest types with high reliability. Performance importance analysis indicates that NDVI plays a dominant role by capturing vegetation structure and seasonal dynamics, while MNDWI contributes to mangrove differentiation through tidal-driven hydrological characteristics despite its lower importance.