Explainable AI Models for Identifying Personality-Driven Risk Factors in Psychosomatic Disorders

Abstract

The objective of this study was to identify and transparently explain personality-driven risk factors associated with psychosomatic symptom severity using explainable artificial intelligence models. This cross-sectional study was conducted on an adult clinical sample recruited from psychosomatic and general health settings in Chile. Participants completed validated self-report instruments assessing personality traits, stress-related psychological variables, and psychosomatic symptom severity. After data preprocessing and standardization, multiple supervised machine learning models were developed to predict psychosomatic symptom severity based on personality and psychological predictors. Model performance was evaluated using cross-validation procedures. The best-performing model was further analyzed using explainable artificial intelligence techniques to identify global and individual-level contributions of predictors, enabling transparent interpretation of nonlinear effects and interactions among personality traits. Nonlinear ensemble models significantly outperformed linear models in predicting psychosomatic symptom severity, explaining a substantial proportion of variance. Stress reactivity emerged as the strongest predictor, followed by emotional instability, perceived stress, and negative affectivity. Explainable analyses revealed threshold and interaction effects, indicating sharp increases in psychosomatic risk at high levels of stress reactivity and compounded effects when combined with emotional instability. Self-regulation-related traits demonstrated a protective effect, particularly at low to moderate stress levels, although this effect diminished under extreme stress conditions. Individual-level explanations highlighted marked heterogeneity in risk profiles, with distinct personality configurations driving symptom severity across participants. The findings indicate that psychosomatic symptom severity is shaped by dynamic, nonlinear interactions among personality traits and stress-related factors rather than isolated linear effects.