Manual and Machine Learning Approaches for Classifying Real and Forged Signatures-A Comparative Study and Forensic...

Area of Science:

• Forensic Science
• Biometrics
• Artificial Intelligence

Background:

• Handwritten signatures serve as unique biometric identifiers for document approval.
• Manual signature authenticity verification is labor-intensive for forensic document examiners.
• Artificial intelligence (AI) offers potential to streamline signature forgery detection.

Purpose of the Study:

• To classify handwritten signatures as genuine or forged using manual methods and machine learning (ML).
• To compare the accuracy and efficiency of ML models against traditional forensic analysis.
• To evaluate the potential of AI in reducing manual workload for signature verification.

Main Methods:

• Collected 1400 signatures (700 genuine, 700 forged) from 71 participants.
• Performed manual examination by comparing genuine and forged signatures.
• Utilized ML models, including Support Vector Machine (SVM) and Random Forest Classifier (RFC), for automated classification.

Main Results:

• Manual examination identified all submitted signatures as imitations.
• Random Forest Classifier (RFC) achieved 92% accuracy in distinguishing genuine from forged signatures.
• Support Vector Machine (SVM) achieved 89.64% accuracy in signature classification.

Conclusions:

• ML models demonstrate high accuracy in classifying handwritten signatures, suggesting their utility in forensic applications.
• AI-based signature analysis can significantly reduce manual effort and increase examination speed for forensic experts.
• Widespread adoption of AI in signature verification faces challenges due to the lack of standardized regulations and universal standards.