A comparison of supervised machine learning models and large language models in predicting personality traits and cognitive ability from asynchronous video interviews

The present manuscript is based on a dataset that contains video interviews (n = 646 participants; k = 3,876 video files; ~ 85 hours of video footage) along with annotations on (a) personality traits (self- and observer reports), (b) cognitive ability (self- and observer reports), and (c) demographic information. The videos contained in the present dataset are based on asynchronous video interviews (AVIs). The manuscript provides a comparison of a supervised machine learning model and two large language models (LLMs; OpenAI, DeepSeek) in predicting personality traits and cognitive ability based on voice characteristics (audio features), facial expressions (visual features), and transcribed text (verbal features). In the case of LLMs, the analysis is based on verbal features, only. The present analysis is meant as a proof-of-concept for a paper that will be submitted for publication in the forthcoming months. In the following pages I provide a summary of the methods and the main results of the statistical analyses.