A Comparative Study of Machine Learning Classification Algorithms for Benthic Habitat Mapping in West Bali National Park Shofura Afanin Nuha (1*), Dita Amelia Pratiwi (1), Alfian(1), and Pramaditya Wicaksono (2)
1 Departement of Geographic Information Science: Postgraduate Student, Faculty of Geography, Universitas Gadjah Mada, Indonesia *shofuraafaninnuha540682[at]mail.ugm.ac.id
2 Departement of Geographic Information Science: Lecturer, Faculty of Geography, Universitas Gadjah Mada, Indonesia
Abstract
The West Bali National Park (Taman Nasional Bali Barat/TNBB) encompasses a wide range of ecosystems, including coastal areas that host diverse habitats such as mangroves, seagrass, and coral reefs. This unique conservation area, located in the Jembrana and Buleleng Districts, spans approximately 3,415 hectares and supports the growth and protection of rich biodiversity, including benthic habitats. Benthic habitats are essential components of coastal ecosystems, comprising seagrass, coral reefs, macroalgae, and various types of substrate. This study aims to compare the performance of two machine learning classification algorithms Random Forest (RF) and Support Vector Machine (SVM) in mapping benthic habitat composition in the Teluk Terima area of TNBB, using 3-meter resolution PlanetScope satellite imagery with visible and near-infrared (NIR) spectral bands. The classification focuses on four primary benthic habitat classes: dominant seagrass, dominant coral, dominant macroalgae, and dominant substrate. The RF algorithm produced an overall accuracy of 59.1%, mapping 27.3 ha of dominant seagrass, 35.9 ha of coral, 0.5 ha of macroalgae, and 40.2 ha of substrate. In comparison, the SVM algorithm resulted in a lower overall accuracy of 47.3%, mapping 35.4 ha of seagrass, 43.9 ha of coral, 15.5 ha of macroalgae, and 9.2 ha of substrate. Accuracy comparisons indicate that RF is more stable in identifying seagrass and substrate classes, while SVM performs better in detecting coral and macroalgae, despite imbalances in user and producer accuracy. These findings suggest that the choice of classification algorithm significantly affects benthic habitat mapping outcomes, and Random Forest offers more consistent results in shallow water environments with complex substrate compositions.
Keywords: Benthic Habitat, Conservation, Random Forest, Support Vector Machine, Machine Learning
