Journal
SPECTROCHIMICA ACTA PART B-ATOMIC SPECTROSCOPY
Volume 123, Issue -, Pages 114-120Publisher
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.sab.2016.08.008
Keywords
Laser-Induced Breakdown Spectroscopy, LIBS; Outlier filtering; Principal Component Analysis, PCA; Linear correlation; Total spectral intensity; Soft Independent Modelling of Class Analogies, SIMCA
Categories
Funding
- Ministry of Education, Youth and Sports of the Czech Republic under the National Sustainability Programme II [LQ1601]
- GACR project: Design of advanced materials using selective laser melting [15-232745]
- European Regional Development Fund [CZ1.05/1.1.00/02.0068]
- project CEITEC Central European Institute of Technology [STI-S-14-2523]
Ask authors/readers for more resources
In this manuscript we highlight the necessity of outlier filtering prior the multivariate classification in Laser-Induced Breakdown Spectroscopy (LIBS) analyses. For the purpose of classification we chose to analyse BAM steel standards that possess similar composition of major and trace elements. To assess the improvement in figures of merit we compared the performance of three outlier filtering approaches (based on Principal Component Analysis, linear correlation and total spectral intensity) already separately discussed in the LIBS literature. The truncated data set was classified using Soft Independent Modelling of Class Analogies (SIMCA). Yielded results showed significant improvement in the performance of multivariate classification coupled to filtered data. The best performance was observed for the total spectral intensity filtering approach gaining the analytical figures of merit (overall accuracy, sensitivity, and specificity) over 98%. It is noteworthy that the results showed relatively low sensitivity and high specificity of the SIMCA algorithm regardless of the presence of outliers in the data sets. Moreover, it was shown that the variance in the data topology of training and testing data sets has a great impact on the consequent data classification. (C) 2016 Elsevier B.V. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available