Classification and identification of soot source with principal component analysis and back-propagation neural network

Abstract

Identification of soot sources is significant in fire investigation and forensic science. In this paper, principal component analysis (PCA) and a back-propagation (BP) neural network model have been used to classify and identify the soot samples from three different kinds of combustible material. Diesel, polystyrene and acrylonitrile butadiene styrene were burnt under the controlled combustion conditions in small-scale burn tests. Based on the matrix data from the GC-MS analysis data, two principal components have been obtained from PCA analysis with the cumulative energy content of 90.21%. Three different kinds of soot sample can be classified with 100% accuracy. A BP neural network model for predicting and identifying the soot source has been further developed. Accurate identification of the unknown samples has been achieved with this trained BP model. This pilot study indicates that PCA and BP neural network methods have potential in the analysis of soot to identify its principle pre-combustion source material. © 2013 Australian Academy of Forensic Sciences.