Ignitable liquid identification using gas chromatography/mass spectrometry data by projected difference resolution mapping and fuzzy rule-building expert system classification Weiying Lu a , J. Graham Rankin b , Alexandria Bondra b,1 , Carolyn Trader b,2 , Amanda Heeren b,3 , Peter de B. Harrington a, * a Center for Intelligent Chemical Instrumentation, Clippinger Laboratories, Department of Chemistry and Biochemistry, OHIO University, Athens, OH 45701-2979, USA b Forensic Science Center, 1401 Forensic Science Drive, Marshall University, Huntington, WV 25701, USA 1. Introduction Ignitable liquid (IL) identification is an important topic in arson crime investigation, because liquids such as gasoline and kerosene are commonly used as accelerants in arson crimes. In this study, automated computational approaches to determine the probable sources of gasoline or kerosene were quantitatively compared. The American Society for Testing and Materials (ASTM) E1618 standard method has been devised to identify different types of ILs by gas chromatography/mass spectrometry (GC/MS) [1]. More specific classification is required in some actual casework. For example, when comparing fire debris samples with an IL sample found in a suspect’s possession or on his clothing, classification of the type of ILs is not specific because of the wide availability of gasoline and kerosene. Assessing IL samples to a potential common source is a complicated problem affected by many parameters. The components of ILs vary due to crude oil source, production processes, and blending at the refinery or in storage tanks at retail outlets. Moreover, different sampling conditions and experimental parameters could greatly influence the IL analysis as well, such as weathering, matrix effects, pyrolysis products from different samples, and extraction and enrichment methods for the samples. Therefore, it is important, however complicated to be able to compare two or more samples in a case to determine if the ignitable liquid residues share a common source. In addition, determination of the precision and error rates for the comparison of IL samples are also important for the evidence to have legal standing in many Daubert states. The recent National Academy of Science report recommends that pattern recognition techniques (of which ignitable liquid residue analysis is one) have established error rates to meet Daubert rules of evidence in court [2,3]. Pattern classification, or classification is an approach to group data or evidence into subsets. Classification is frequently used in forensic science to place a sample or evidence into a known subsets that may be used to achieve individualization of a common source. Modern chemical instruments readily furnish digitized chemi- cal data or a chemical fingerprint. These data are amenable to a variety of computational methods for classification. These computational classification methods are generally referred to as pattern recognition. Henceforth, classification refers to the computational classification of chemical data by source (i.e., sample). The process of validation is used to measure the statistical error rates of the classifiers. A fundamental validation method is accomplished by applying the computational classifier or model to an independent set of data of known origin or source. The Forensic Science International 220 (2012) 210–218 A R T I C L E I N F O Article history: Received 23 June 2011 Received in revised form 29 February 2012 Accepted 2 March 2012 Available online 1 April 2012 Keywords: Ignitable liquid identification Gas chromatography/mass spectrometry Fuzzy rule-building expert system Projected difference resolution mapping A B S T R A C T The gasoline and kerosene collected from different locations in the United States were identified by gas chromatography/mass spectrometry (GC/MS) followed by chemometric analysis. Classifications based on two-way profiles and target component ratios were compared. The projected difference resolution (PDR) mapping was applied to measure the differences among the ignitable liquid (IL) samples by their GC/MS profiles quantitatively. Fuzzy rule-building expert systems (FuRESs) were applied to classify individual ILs. The FuRES models yielded correct classification rates greater than 90% for discriminating between samples. PDR mapping, a new method for characterizing complex data sets was consistent with the FuRES classification result. ß 2012 Elsevier Ireland Ltd. All rights reserved. * Corresponding author. Tel.: +1 740 994 0265. E-mail address: Peter.Harrington@OHIO.edu (P.d.B. Harrington). 1 Currently at El Paso County Sheriff’s Office, El Paso, TX, USA. 2 Currently at Eastern Kentucky Forensic Laboratory, Ashland, KY, USA. 3 Currently at Wyoming State Crime Laboratory, Cheyenne, WY. Contents lists available at SciVerse ScienceDirect Forensic Science International jou r nal h o mep age: w ww.els evier .co m/lo c ate/fo r sc iin t 0379-0738/$ – see front matter ß 2012 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.forsciint.2012.03.003