Partial least squares discriminant analysis: a dimensionality reduction method to classify hyperspectral data

Abstract

The recent development of more sophisticated spectroscopic methods allows acquisition of high dimensional datasets from which valuable information may be extracted using multivariate statistical analyses, such as dimensionality reduction and automatic classification (supervised and unsupervised). In this work, a supervised classification through a partial least squares discriminant analysis (PLS-DA) is performed on the hyperspectral data. The obtained results are compared with those obtained by the other discrimination functions and the commonly used classification approaches. The results show that PLS-DA, using three components explains the 97% of the total variance in the data, and obtains a better-defined partition than other discussed approaches).