The prediction and impact of self-criticism on non-suicidal self-injury in adolescents: A study based on machine...

This study uniquely combines machine learning with structural equation modeling to investigate how self-criticism affects non-suicidal self-injury (NSSI) behaviors in adolescents. Using a Support Vector Machine (SVM) for binary classification (distinguishing NSSI presence/absence) and XGBRegressor for regression (predicting NSSI severity), with self-criticism, emotional regulation difficulties, and public self-consciousness as key predictors, the study accurately classified and forecasted NSSI behaviors. The K-fold cross-validation confirmed the robustness of these predictors. The SVM classifier achieved an accuracy of 85.3% (AUC = 0.89, 95% CI [0.87, 0.91]), while the regression model explained 50.9% of variance (R² = 0.509). Structural equation modeling revealed that self-criticism directly influences NSSI (β = 0.32, p < .001) and indirectly affects it through both emotional regulation difficulties (β = 0.18, 95% CI [0.14, 0.22]) and public self-consciousness (β = -0.11, 95% CI [-0.15, -0.07]), with a significant serial mediation pathway (self-criticism → emotional regulation difficulties → public self-consciousness → NSSI). These findings demonstrate that self-criticism, difficulty in emotional regulation, and public self-consciousness can predict NSSI risk/severity and differentiate between groups with and without NSSI behaviors in our sample. Additionally, adolescents' public self-consciousness can directly and negatively predict non-suicidal self-injurious behaviour, revealing its protective role in the pattern of NSSI behaviour and providing support for the assumptions in the benefit and hindrance models of NSSI.