Predicting honest behavior based on Eysenck personality traits and gender: an explainable machine learning study using SHAP analysis.

Introduction: This study bridges a critical gap in aviation safety research by examining how Eysenck personality traits (Neuroticism, Psychoticism, Extraversion) and gender predict dishonest behavior in high-risk aviation contexts. While prior studies have focused on the Big Five and HEXACO models in ethical decision-making, empirical applications of the Eysenck framework to honesty prediction remain scarce-particularly in aviation, where dishonest acts (e.g., underreporting safety incidents) carry severe public safety consequences. Methods: We collected behavioral data from 102 flight and air traffic control cadets using a coin-toss task. Explainable machine learning (XGBoost) was employed to model nonlinear relationships between personality, gender, and honesty. Model performance was evaluated via AUC, with SHAP analysis identifying key predictors. Results: XGBoost achieved superior predictive accuracy (AUC = 0.802). SHAP analysis revealed: (1) neuroticism as the strongest predictor of dishonesty; (2) significant gender differences (higher dishonesty rates in males); and (3) threshold effects for Psychoticism and Extraversion. Discussion: This work makes three key contributions: (1) first systematic application of the Eysenck model to aviation honesty prediction; (2) identification of gender as a critical moderating variable; and (3) development of a SHAP-driven interpretable framework that connects machine learning outputs with psychological theory. Practically, these findings enable data-driven screening of cadets' honesty tendencies during recruitment, facilitating targeted interventions for safer aviation operations.