Hyperspectral image (HSI) classification is a very important subject in the area of remote sensing. In common, the intricate distinctiveness of hyperspectral data makes the precise classification of such data difficult for traditional machine learning methods. In addition, hyperspectral imagery often deals with an innately nonlinear relationship between the spectral information captured and therefore the corresponding materials. Machine learning has been acknowledged as a robust feature extraction tool to successfully address nonlinear issues and widely used in image processing tasks. This survey paper presents a scientific review of machine learning-based HSI classification using Graph-cut and Local Covariance Matrix Representation (LCMR) method and compares various strategies for this topic. Specifically, we first summarize the foremost challenges of HSI classification which cannot be effectively overcome by traditional methods, and also introduce the advantages of Support Vector Machine (SVM) model to handle these problems. Experimental results have been conducted using publicly available hyperspectral data sets for classiﬁcation.