Machine learning methods for credibility assessment of interviewees based on posturographic data.

Area of Science:

• Biomedical Engineering
• Forensic Science
• Human-Computer Interaction

Background:

• Traditional credibility assessment methods can be invasive or unreliable.
• Posturography, the analysis of body sway, offers a potential non-invasive physiological measure.
• Developing objective measures for deception detection is a significant challenge.

Purpose of the Study:

• To investigate the efficacy of posturographic signals for non-invasive credibility assessment.
• To explore the feasibility of an autonomous deception detection system using machine learning.
• To identify reliable features from center of pressure (COP) signals indicative of deception.

Main Methods:

• An interview paradigm was used, recording center of pressure (COP) signals from force plates.
• Subjects were instructed to respond with either truthful or deceptive intent.
• Machine learning classification models were trained and evaluated using COP-derived features.

Main Results:

• The proposed method demonstrated high efficiency and non-invasiveness.
• Classification models achieved a best accuracy of 93.5% for detecting deceptive responses.
• Feature sets derived from COP signals were effective in distinguishing truthful from deceptive statements.

Conclusions:

• Posturographic signals from force plates are a promising tool for non-invasive credibility assessment.
• Machine learning algorithms can be effectively employed to build autonomous deception detection systems.
• This approach offers a novel, objective, and efficient method for credibility evaluation.