Accurate terrain classification is important in the navigation and decision-making of unmanned ground vehicles (UGVs) in unstructured environments. This work makes use of the Multi-Spectral Imaging (MSI) datasets with 9, 16, and 25 spectral bands to compare the performance of three machine learning models: Random Forest (RF), k-Nearest Neighbors (kNN), and Support Vector Machine (SVM). The experimental results indicate that the highest accuracy of 87.03% is achieved by using 25 spectral bands, while RF is robust to noisy datasets. kNN outperformed in terms of precision and recall for certain terrain categories, while SVM offered the worst results with computational complexity and class imbalance. The obtained results stress the relevance of spectral band selection for improvement in classification accuracy.