Detecting Young Children's Deception Using Facial Expression Analysis With High Dimensional Statistical Methods

Area of Science:

• Developmental Psychology
• Computational Social Science
• Affective Computing

Background:

• Deception is prevalent in early childhood, yet accurate detection remains difficult.
• Traditional methods often fail to capture the dynamic nature of children's facial expressions during deception.

Purpose of the Study:

• To assess the feasibility of using high-dimensional statistical methods to detect deception in children via facial expressions.
• To investigate the impact of incorporating time-dependent facial expression statistics (slope and fitness) on deception detection accuracy.

Main Methods:

• Analyzed facial expressions of 93 children (3-6 years old) during a guessing game using a machine vision algorithm (FACET).
• Extracted mean, slope, and fitness statistics for nine basic facial expressions.
• Employed high-dimensional feature selection and compared logistic regression, linear discriminant analysis, random forest, and adaptive boosting models.

Main Results:

• Incorporating time-dependent statistics (slope and fitness) improved average balanced accuracy from 47.2% to 62.0%.
• Feature selection further enhanced model performance, increasing balanced accuracy from 60.8% to 63.3%.
• Balanced accuracy proved a more reliable performance measure than original accuracy (61.5% vs. 54.6%).

Conclusions:

• Dynamic facial emotional features, analyzed with computational methods, show promise for detecting deception in early childhood.
• Time-dependent statistical analysis significantly enhances the accuracy of deception detection models.
• This approach offers potential for advancing research in child development, emotion, and social cognition.