Hyperspectral imaging has been demonstrated to possess a high discrimination performance and is currently employed in a multitude of applications in the field of resource development. However, the data structure is intricate, and machine learning is indispensable for data processing and analysis. In this study, we applied 10 machine learning models (CNN, Tree, Discriminant, Logistic Regression, Naïve Bayes, Support Vector Machine, KNN, Ensemble, (shallow) Neural Network, and Kernel Approximation) to four major hyperspectral datasets commonly used as benchmarks (Indian Pines, Salinas, Pavia University, and Botswana) to compare their classification performance. The results indicated that the Support Vector Machine (SVM) demonstrated the highest performance on the Indian Pines dataset, whereas the Neural Network exhibited the best performance across the other datasets. In contrast, CNN showed moderate performance on these datasets. These findings suggest the necessity of selecting appropriate machine-learning models based on the specific characteristics of the datasets. In general, newer machine learning models such as CNNs tend to have higher computational costs and pose challenges in terms of interpretability. This study demonstrated that classical machine learning models can, in some cases, outperform state-of-the-art models, highlighting the potential of traditional approaches in achieving superior performance under certain conditions.