A multimodal approach to faking detection in a selection interview

Abstract

The aim of this study was to investigate the possibility of faking detection in a selection interview using a multimodal approach based on paraverbal, verbal/nonverbal cues, and facial expressions. In addition, we compared detection accuracies of simple linear and complex nonlinear machine learning algorithms. A sample of 102 participants were interviewed in two conditions—honest responding and simulated highly realistic selection. Results showed only several significant univariate effects of experimental condition for paraverbal, verbal, and facial expression cues. All the algorithms performed comparably and above chance levels, except for random forests, which overfitted on the training sets and underperformed on the testing sets. Still, considering the algorithms' accuracy was limited, usefulness of multimodal data for deception detection remains questionable.