Manual and Machine Learning Approaches for Classifying Real and Forged Signatures—A Comparative Study and Forensic Implications

ABSTRACTA handwritten signature is one of the forms of a biometric measure that creates an individual identity of the persons to mark their approval related to any document. The manual examination for determination of the authenticity of the handwritten signatures is a common practice amongst forensic document examiners. This process involves a detailed and thorough analysis of handwriting characteristics of an individual ensuring a comprehensive assessment of the each and every important feature. However, the use of artificial intelligence tools can reduce this manual work of experts for identifying forgery in signatures. The main objective of the present study was to classify the handwritten signatures as forged and genuine, manually as well as using tools of artificial intelligence, especially machine learning (ML) methods. A total of 1400 signatures, consisting of 700 forged and 700 real signatures were obtained. The signatures were obtained from 71 participants; one writer executed 700 signatures (real/genuine signatures) and 70 participants were asked to forge 10 signatures each by observing one genuine signature selected from a pool of 700 real signatures. The study employed two methods to examine the signatures: manual examination and by using machine learning‐based models. In the manual examination, thorough comparison between real and forged signatures revealed that all the forged signatures were imitated and falsified that is not created by the original creator. In contrast, the machine learning‐based models that is support vector machine (SVM) and random forest classifier (RFC) were utilized for classifying the signatures as either forged or genuine. The RFC and SVM achieved accuracies of 92% and 89.64% respectively for classification of the signatures as real or forged. Accuracy of both the models of the machine learning approach revealed that the approach may be used to reduce the manual work of forensic handwriting experts and allow this examination to be performed more quickly. However, the admissibility of AI‐based examination of signatures is still challenged due to the lack of universal standards and a regulatory framework.