Machine Learning Assisted LIBS Classification of Burnt and Unburnt Paper Samples: A Forensic Perspective

Abstract

A crucial part of investigating into a possible arson case is analyzing the chemical makeup of the fire's effects. In order to identify the chemical makeup of burnt materials that are frequently found in homes and businesses, the current study attempts to explore the potential of LIBS in forensic applications including burning of as well as forgery in paper-based documents. Various kinds of paper were chosen as a model sample to investigate the potential of Laser induced breakdown spectroscopy (LIBS) in forensics. First of all, LIBS spectra were recorded for unburnt samples by using Nd: YAG laser operated at 532nm, by optimizing the specific energy and delay between the laser pulse and shutter of the spectrometer, their elemental compositions were evaluated. The plasma parameters such as, plasma temperature, and electron number density were studied comprehensively to validate the local thermodynamic equilibrium condition. The three different kinds of ignition sources namely disposable lighter, candle, and gas stove were used to burn the sample to check the variation in elemental composition after the burning process for forensic applications as these sources are often present in homes and workplaces. The LIBS results suggest that all the samples have almost the same elemental composition even after using various kinds of ignition sources. Therefore, machine learning was applied to LIBS data by recording 100 datasets for each sample when ignited with different sources. PCA discriminates the samples slightly, but supervised machine learning algorithms LDA, QDA, and Linear SVM showed superior i.e., 100 % classification accuracies for various datasets that suggest the machine learning-assisted LIBS is a promising tool for forensic applications.