NCJ Number
225887
Journal
Journal of Forensic Sciences Volume: 54 Issue: 1 Dated: January 2009 Pages: 49-59
Date Published
January 2009
Length
11 pages
Annotation
This study assessed the efficiency of likelihood ratio (LR)-based measures in their application to solving various classification problems for glass objects with respect to elemental composition and refractive index (RI), and these measures were compared to other classification methods.
Abstract
The study found that the proposed scheme for classifying class fragments works with efficiency, with the exception of special care used in classifying car windows and building windows, which are similar in elemental content. The classification of these categories of glass should focus on the combination of elemental content and information on the change in refractive index during annealing. The application of support vector machines (SVM) and naïve Bayes classifiers (NBC) produced slightly better results than the LR model; however, the observed differences in misclassification rates were not great, and no single classification method was clearly more effective than the others. The LR model is recommended, since this framework has the advantages of ease of interpretation, and it does not act as a “black-box,” unlike the SVM method; and it does not require the investigator to make an assumption about prior belief, unlike the NBC method. In addition, the LR model might be easily adapted to other forensic classification problems in which the observation may be multivariate. A total of 153 class objects (23 building windows, 25 bulbs, 32 car windows, 57 containers, and 16 headlamps) were analyzed by scanning electron microscopy coupled with an energy dispersive X-ray spectrometer. Refractive indexes for building and car windows were measured before and after an annealing process. 7 tables, 4 figures, and 35 references