Impact of Laser-Induced Breakdown Spectroscopy data normalization on multivariate classification accuracy

Abstract

Multivariate data analysis (MVDA) is getting popular across the spectroscopic community. To assess accurate results, the obtained data should be preprocessed prior to utilization of any MVDA algorithm. The process of data normalization or “internal standardization” is widely used across a broad range of applications. In this manuscript we investigate the utilization of Laser-Induced Breakdown Spectroscopy (LIBS) coupled with MVDA. However, many articles regarding the use of MVDA on data from LIBS do not provide any information about the data pretreatment. This work describes the impact of LIBS data normalization approaches on MVDA classification accuracy. Also, the impact of classical data preprocessing (mean centering and scaling) exploiting the prior utilization of MVDA was studied. This issue was investigated exploiting simple soft independent modelling of class analogies algorithm. The findings were generalized for three sample matrices (steel, Al alloys, and sedimentary ores). Furthermore, the selection of an appropriate normalization algorithm is not trivial since the spectrum of each sample matrix is composed of a different number of elements and corresponding elemental lines.