Detection of simple and complex deceits through facial micro-expressions: a comparison between human beings’ performances and machine learning techniques

Micro-expressions have gained increasing interest in the last few years, both in scientific and professional contexts. Theoretically, their emergence suggests ongoing concealments, making them arguably one of the most reliable cues for lie detection (e.g., Yan, Wang, Liu, Wu & Fu, 2014; Venkatesh, Ramachandra & Bours, 2019). Given their fast onset, they result almost imperceptible to the eye of an untrained subject, making it necessary to work on automatic detection tools. Machine learning models have shown promisingly results within this domain; thus, the aim of the study at hand, was to compare the performances human judges and machine learning models obtain on the same dataset of stimuli. Regrettably, machine learning performances have ended up being around the chance level, positing the question of why previous and a-like studies have collected better results. Briefly, insights on how to properly organize an experimental paradigm and collect a dataset for lie detection studies are discussed, while concluding that among other several necessary cues it is still crucial to consider micro-expressions when dealing with lie detection procedures.