Moment Projection Analysis as New Clustering Method

Abstract

New clustering algorithm, so called Moment Projection Analysis (MPA) has been developed and its performance has been evaluated for three separate vibrational spectroscopic datasets (IR, NIR and Raman). The first step in MPA is the calculation of moment of each spectrum. Then, the moment was projected onto two dimensional feature space using the project vector by maximizing the Fisher`s information criteria. Consequently, each spectrum was represented as the new coordinates in two dimensional feature space and the coordinates were used for clustering. Simultaneously, Principal Component Analysis (PCA) was also accomplished to compare the performance of MPA. Three datasets of IR spectra of colon adenoma samples, NIR spectra of petroleum products and Raman spectra of ginseng were used. In overall, MPA showed the improved discrimination capability over PCA especially when spectral features were close each other.