Building a second-opinion tool for classical polygraph.

Area of Science:

• Forensic Science
• Artificial Intelligence
• Psychology

Background:

• Polygraph screenings are vital in sectors like law enforcement and finance.
• A significant concern is the inherent error rate in polygraph testing.
• Human error by polygraph examiners contributes substantially to screening inaccuracies.

Purpose of the Study:

• To apply machine learning (ML) for detecting human errors in polygraph examiner conclusions.
• To develop and validate a "second-opinion" tool to mitigate subjectivity in polygraph assessments.
• To identify novel features that enhance the accuracy of error detection models.

Main Methods:

• Training an ML error detection model without pre-labeled errors.
• Implementing and testing a practical "second-opinion" tool for examiner error identification.
• Conducting experiments to analyze deception patterns across different topics.

Main Results:

• Successful development of an ML-based tool to detect examiner errors in polygraph screenings.
• Demonstrated reduction in subjectivity associated with polygraph assessments.
• Identification of key features that significantly improve model accuracy.

Conclusions:

• Machine learning offers a viable approach to enhance the reliability of polygraph screenings.
• The developed tool can serve as a valuable aid in identifying human errors, improving objective decision-making.
• Findings pave the way for a re-evaluation of traditional polygraph practices.