Problems of Forensic Sciences 2020 Vol. 124 (CXXIV) 237-254
THE POTENTIAL USE OF PRINCIPAL COMPONENT ANALYSIS AND MAHALANOBIS DISTANCE IN HANDWRITING COMPARISON
Loren WILLIAMS1, Morgan MILLS2, Carol RITTER1, Lawrence QUARINO1
1Forensic Science Program, Cedar Crest College, Allentown, PA, USA
2Bode Technology, Lorton, VA, USA
Streszczenie
The potential use of multivariate statistical analysis in handwriting comparison to provide objective criteria to an inherently subjective discipline was explored in this study. Twenty sets of exemplar writing in the form of checks from fifty participants were used as a primary dataset. Using 3-dimensional principal component analysis (PCA) plots of the first three principal components, writing from thirty-eight of the fifty participants could be differentiated. Applying both PCA and Mahalanobis distance to unknowns created from a closed set of exemplar writing from ten participants in the dataset resulted in a true positive rate of 70%. Conversely, the use of PCA and Mahalanobis distance in the comparison of seven true negatives to the entire database resulted in a false positive rate of only 1.1%. The method was unsuccessful in helping to determine the possible origin of both traced and simulated writing, although forged writing could be differentiated from exemplar writing. The method was also able to correctly determine that the writing on two checks from a group of six checks had a common origin even when a professional forensic handwriting examiner erroneously concluded that no common origin existed. The paper does not advocate for a replacement of traditional handwriting examination with objective criteria such as the methodology described in the study. However, the results of the study support the incorporation of mathematical models into standard practice of handwriting comparison as a means of verification and as a way to ensure consistency of conclusions.
Słowa kluczowe
Handwriting comparison; Principal Component Analysis; Mahalanobis distance.