Machine Learning Techniques for Geochemical Analysis Using Laser-Induced Breakdown Spectroscopy

• Published in: Applied spectroscopy Vol. 79; no. 7; p. 1129
• Main Authors: Akbar, Shamaila, Razzaq, M Inzmam, Ahmed, Nasar, Abbas, Kamran, Rafique, M, Baig, M Aslam, Hedwig, Rinda, Farooq, Zahid
• Format: Journal Article
• Language: English
• Published: United States 01.07.2025
• Subjects: geochemical analysis; analysis of variance; standard normal variate; ANOVA; Laser-induced breakdown spectroscopy; SVM; machine learning techniques; LIBS; principal component analysis; PCA
• ISSN: 1943-3530
• DOI: 10.1177/00037028251334151

In the present work, appropriate machine learning techniques coupled with LIBS have been proposed for the effective classification of multielement rock samples. To obtain the best classification efficiency most suitable emission lines were selected. Plasma on the surface of seventeen rock samples was generated using a 532  nm Q-switched neodymium-doped yttrium aluminum garnet (Nd:YAG) laser, and optical emission spectra were collected via an Avantes spectrometer. Well-isolated signature emission lines corresponding to detected elements (Ca, Mg, Na, K, Fe, Ba, Sr, Si, Al, and Li) were chosen as input for the machine learning algorithms. Three machine learning techniques, including analysis of variance (ANOVA), principal component analysis (PCA), and PCA coupled with standard normal variate (SVM), were utilized on normalized intensities of selected spectral lines of detected elements. ANOVA testing on the selected lines was employed to assess the normality and suitability of data for further machine learning techniques. The combination of laser-induced breakdown spectroscopy (LIBS) with PCA enabled a comprehensive classification of rock samples. The linearity and efficiency of PCA were enhanced by utilizing the support vector machine (SVM), resulting in the accurate classification of rock samples. This study demonstrates that to assess the effective classification of multielement rock samples the appropriate emission lines and machine learning techniques are crucial. Using this methodology results become more reliable as compared to conventional machine learning techniques.