RETRACTED: Digital Phenotyping of Smartphone Data Successfully Predicts a Broad Range of Personality Constructs

Abstract

Digital Phenotyping (DP) entails exploring digital expressions of human personality using behavioral cues drawn from smartphones’ digital footprints. Most personality-oriented DP studies focus narrowly on the Big5 model. This research aims to broaden this approach, using fifty-four personality constructs rooted in fifteen leading personality theories beyond the Big5. Our sample consisted of 104 respondents from whom smartphone data was collected over 7–10 days. We implemented both deductive (hypothesis-testing) and inductive (machine learning) modelling methods. Results show that fifteen of the sixteen broad personality constructs were successfully predicted from smartphone data (forty-eight sub-personality items of the fifty-nine types and personality traits). The best overall predictive model was Gradient Boosted Trees with communication-related features having the highest predictive weight. DP has the potential to transform the field of personality research and may be applied in areas such as HR analytics, personality-based targeted marketing, individualized homeland security, financial risk assessments, personalized medicine, and more.