Using Artificial Intelligence to Detect Faking Good in a Big Five Personality Self-Report

The current study was designed to develop a supervised machine learning classifier to identify faking good by analyzing items response patterns of a Big Five Personality self-report. In our between-subject design, we divided participants (N=239) into two groups and directly manipulated their faking behaviors via different instructions sets. We implemented three models to classify participants as being honest or fake respondents: (i) a simple classifier based on the self-report Lie scale as cut-off score (CBC); (ii) a logistic regression with the personality and control scales as predictors (LRC); (iii) a machine learning classifier fitted directly on the personality items (XGB). The machine learning model of our choice was XGBoost which belongs to a subset of supervised algorithms called ensemble methods. In confirmation of our hypothesis, XGB prediction accuracy was significantly higher than that of CBC and LRC.