Accurate and efficient tree species classification and mapping is crucial for forest management and conservation, especially on the Loess Plateau, where forest quality urgently needs improvement. This study selected three research sites-Yongshou (YS), Zhengning (ZN), and Yanchang (YC)-on the Loess Plateau and classified the main forest tree species using RGB images acquired by an unmanned aerial vehicle (UAV). The RGB images were normalized, and vegetation indices (VIs) were extracted. Feature selection was performed using the Boruta algorithm. Two classifiers, Support Vector Machine (SVM) and Random Forest (RF), were used to evaluate the contribution of different input features to classification and their performance differences across regions. The results showed that YC achieved the best classification performance with an overall accuracy (OA) of over 83% and a Kappa value of at least 0.78. The results showed that YC achieved the best classification performance (OA > 83%, Kappa >= 0.78), followed by ZN and YS. The addition of VIs significantly improved classification accuracy, particularly in the YS region with imbalanced sample distribution. The OA increased by more than 13.27%, and the Kappa improved by more than 0.17. Feature selection retained most of the advantages of the complete feature set, achieving slightly lower accuracy. Both RF and SVM are effective for tree species classification based on RGB images, with comparable performance (OA difference <= 1.5%, Kappa difference < 0.02). This study demonstrates the feasibility of UAV-based RGB images in tree species classification on the Loess Plateau and the great potential of RGBVIs in tree species classification, especially in areas with imbalanced class distributions. It provides a viable approach and methodology for tree species classification based on RGB images.