Comparative Analysis of Machine Learning Models for Sentinel-2 based Classification of the Bornean Heath Forest

Bornean heath forests, known as hutan kerangas, are fragile ecosystems that face significant anthropogenic threats. This study integrates Sentinel-2 satellite imagery with Machine Learning (ML) models to accurately classify these forests and assess their current spatial distribution. The Random Forest (RF) and Gradient Tree Boost (GTB) models achieved the highest classification performance, with overall accuracy scores of 96.66% and 96.69%, respectively, and Kappa coefficients of 0.945. These metrics were obtained using a test dataset with an 80:20 train-test split and validated through a 5-fold cross-validation process, ensuring the robustness of the models. Compared to previous studies employing unsupervised classification with Landsat-9 data, this approach demonstrates improved classification reliability and spatial accuracy. The findings highlight the substantial potential of combining remote sensing technologies with advanced ML techniques for large-scale ecosystem monitoring. This approach provides valuable insights for conservation planning and sustainable management of Bornean heath forests, addressing the growing environmental pressures that threaten their integrity.