Development of a Deep-Learning-Based Computerized Scoring Algorithm.

Area of Science:

• Forensic Science
• Biomedical Engineering
• Artificial Intelligence

Background:

• Polygraph tests traditionally rely on human examiners to interpret physiological responses, which is prone to subjective biases and errors.
• Existing computerized scoring systems (CSSs) often use linear classifiers, failing to capture the complex, nonlinear nature of biological signals.
• Human biases (political, regional, religious) and examiner fatigue/stress can compromise polygraph accuracy.

Purpose of the Study:

• To develop a Korean computerized scoring system (CSS) that mitigates subjective examiner bias in polygraph analysis.
• To enhance the accuracy of polygraph deception detection by effectively analyzing nonlinear bio-signals.
• To improve upon conventional CSS models that struggle with the inherent complexity of physiological data.

Main Methods:

• Development of a novel Korean computerized scoring system (CSS) employing deep neural networks.
• The system is designed to automatically analyze polygraph charts, focusing on the nonlinear characteristics of bio-signals.
• Performance evaluation using standard metrics: recall, precision, and F1 score.

Main Results:

• The developed deep learning-based CSS achieved high performance metrics: recall (0.9681 ± 0.0314), precision (0.9700 ± 0.0321), and F1 score (0.9683 ± 0.0171).
• Demonstrated significant improvement compared to conventional CSS models that rely on linear classifiers.
• The system effectively addresses the nonlinearity of bio-signals, leading to more accurate deception detection.

Conclusions:

• The proposed Korean CSS, leveraging deep neural networks, offers a substantial advancement in objective and accurate polygraph analysis.
• This approach effectively reduces human error and subjective bias inherent in traditional polygraph scoring.
• The findings support the integration of deep learning for enhanced performance in forensic physiological measurement analysis.