AUTOMOBILISTIC GLASSES AS CRIME SCENE TRACES: A MULTIVARIATE APPROACH

Glasses are common trace evidence elements in crime scenes, and the analysis of this material can be essential for evaluating different criminal dynamics. This work aimed to analyze the possibility of differentiating and classifying windshield glass using multivariate analysis methods. Automotive glass fragments from different vehicle brands were evaluated according to internal and external faces. We have collected from literature EDXRF (Energy Dispersive X-ray Fluorescence) data for different oxides concentrations. These data were organized in a matrix with 56 samples and nine variables. We applied unsupervised (PCA, Principal Component Analysis) and supervised (SIMCA, Soft Interclass Modeling Classification Analogy) methods. We assessed the classification responses through ROC (Receiver Operating Characteristics). As a result, the PCA indicated the presence of two groups of glasses in three main components. SIMCA verified the unsupervised classification, and the distances and interclass residues parameters were adequate with no outliers. The ROC analysis indicated a sensitivity of 0.793, a specificity of 0.815, and an efficiency of 0.804 for predictions. We concluded that multivariate analysis was successful in discriminating between the internal and external faces of automotive glasses.

Keywords:
glass; multivariate analysis; forensic sciences; crime scene