Evaluating Machine Learning Algorithms for Forest Cover Extraction in Kailali, Nepal

Abstract

Forest cover mapping plays a critical role in environmental monitoring, biodiversity conservation, and sustainable land-use planning, especially in ecologically diverse regions like Nepal. This study evaluates the performance of ten supervised machine learning classifiers for forest cover extraction in the Kailali District using Sentinel-2 satellite imagery. The classifiers assessed include Random Forest, Support Vector Classifier, Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Decision Tree, Gaussian Naïve Bayes, AdaBoost, Quadratic Discriminant Analysis, and Gaussian Process Classifier. Feature engineering involved the derivation of 17 vegetation and water indices alongside key spectral bands, followed by correlation analysis to optimize input variables. Ground truth data were collected through field surveys and high-resolution imagery to ensure accurate model training and validation. Classifier performance was evaluated using k-fold cross-validation and standard metrics, including accuracy, precision, recall, and F1-score. Among the models, Random Forest and Gaussian Process achieved the highest classification accuracies of 91.37% and 91.31%, respectively. The study demonstrates the effectiveness of machine learning techniques in forest cover classification and provides valuable insights for enhancing remote sensing-based monitoring frameworks in support of sustainable forest management in Nepal.