Addressing Forest Management Challenges by Refining Tree Cover Type Classification with Machine Learning Models

Abstract

The goals of this paper were twofold: to continue and refine previous research in the topic of tree cover type classification by harnessing modern machine learning models, and to extend the conclusions of that work to demonstrate that results gained from such models can be used to assist U.S. land management agencies in current challenges they face. Using the same dataset as the past study, an artificial neural network was constructed and compared with three baseline traditional machine learning models: Naïve Bayes, Decision Tree, and K-Nearest Neighbor. The artificial neural network achieved 97.01% ac-curacy while the best-performing traditional classifier, K-Nearest Neighbor, managed 74.61%. This mirrored the earlier results, but with higher overall accuracy on both counts. Specifically, the neural network performed 26.43% better than before, showing not only advances in machine learning algorithms over the past 18 years, but also that accuracy is now high enough to apply practically to land management issues where natural resource inventory is time-consuming and expensive.