IMPROVING SPECTRAL ANALYSIS WITH THE APPLICATION OF MACHINE LEARNING: STUDY OF LASER-INDUCED BREAKDOWN SPECTROSCOPY (LIBS) AND RAMAN SPECTROSCOPY WITH CLASSIFICATION AND CLUSTERING TECHNIQUES.

Author

Christopher Mandrell, Southern Illinois University CarbondaleFollow

Date of Award

5-1-2020

Degree Name

Master of Science

Department

Physics

First Advisor

Sivakumar, Poopalasingam

Abstract

AN ABSTRACT OF THE THESIS OFChristopher T. Mandrell, for the Master of Science degree in Physics, presented on April 8, 2020, at Southern Illinois University Carbondale.TITLE: IMPROVING SPECTRAL ANALYSIS WITH THE APPLICATION OF MACHINE LEARNING: STUDY OF LASER-INDUCED BREAKDOWN SPECTROSCOPY (LIBS) AND RAMAN SPECTROSCOPY WITH CLASSIFICATION AND CLUSTERING TECHNIQUESMAJOR PROFESSOR: Dr. Poopalasingam SivakumarAtomic and molecular spectroscopy, in the form of LIBS emissions and Raman scattering, respectively, are tools that provide a vast amount of information with little to no sample preparation. For this reason, these techniques are finding their way into a wide range of fields. However, each spectrum is notoriously complicated to analyze, with many complex interactions at play. Machine learning is the result of work on artificial intelligence. It provides tools to train a computer to look for connections in complex data sets that would likely be missed, or not even looked for, by other analytical methods. The combination of highly informative yet complex data with an analysis that is specifically designed to probe highly complex data for meaningful information is a logical step in the analysis of these spectra. Here we apply statistical analysis and classification algorithms to Raman spectra of pancreatic cancer cells and clustering algorithms to LIBS spectra of Mars Curiosity Rover simulants and Raman spectra of Mars Perseverance Rover simulants. We report here high accuracy in the classification of different types of pancreatic cancer cells, and informative clustering of the two Mars rovers’ simulant data.