Multivariate chemometrics for the forensic discrimination of blue ball-point pen inks based on their Vis spectra.

Fifty blue ball-point pen inks of five different brands were examined on the basis of the Vis spectrum of their ethanolic solutions with a view to achieving good discrimination between them. Samples were dissolved in absolute ethanol and their absorbance values in the range of 400-750 nm, after appropriate transformations, were used as variables in the multivariate statistical techniques of cluster analysis (CA), principal component analysis (PCA) and discriminant analysis (DA). These techniques were used successively so that an effective and meaningful discriminant model was calculated in the final step. The initial 351 variables (log10 transformed ink absorption values at each wavelength) were subjected to a K-means CA over the objects (samples) and only 20 variables were retained. Principal component analysis was used to detect any outliers (four samples were removed) and the remaining samples were re-subjected to PCA to decide how many variables to enter into DA and whether original variables or components should be used. It was found that the first three principal components (in accordance with the Kaiser criterion) were good descriptors of the 20 original variables (96.97% of the data variance was explained) and their use as latent variables in DA lead to low average variable redundancy (33.6%) in the discriminant model. The calculated model had a Wilks' lambda of 8.98 x 10(-5) and was statistically significant at the P = 0.05 level. The post hoc classification of the training dataset was 100% correct. From the DA results and the component loadings it was found that discrimination was achieved on the basis of differences in the shape of the absorption bands as well as their relative intensities. The method was therefore deemed appropriate for supporting exclusionary forensic purposes.