Explainable Boosting Machine Prediction of Emotional Numbness from Trauma Exposure, Dissociation, and Alexithymia
Objective: This study aimed to predict emotional numbness from trauma exposure, dissociation, and alexithymia among trauma-exposed adults in Canada using an interpretable Explainable Boosting Machine model.
Methods and Materials: This quantitative cross-sectional predictive study was conducted on 426 trauma-exposed adults residing in Canada. Participants completed standardized self-report measures assessing emotional numbness, lifetime trauma exposure, dissociative experiences, and alexithymia. Emotional numbness was considered the outcome variable, while trauma exposure, dissociation, and alexithymia were entered as predictor variables. Data were screened for missing values, response validity, outliers, and distributional properties. Descriptive statistics, reliability coefficients, and Pearson correlations were calculated before predictive modeling. The dataset was divided into training, validation, and test subsets. An Explainable Boosting Machine regression model was trained and evaluated against a conventional multiple linear regression model using root mean squared error, mean absolute error, and coefficient of determination. Global and local interpretability outputs were examined to determine predictor importance, non-linear effects, and interaction patterns.
Findings: Inferential analyses showed that trauma exposure, dissociation, and alexithymia were all significantly and positively correlated with emotional numbness. Dissociation had the strongest correlation with emotional numbness, followed by alexithymia and trauma exposure. The Explainable Boosting Machine model outperformed multiple linear regression across predictive indices. In the independent test set, the linear regression model explained 44% of the variance in emotional numbness, whereas the Explainable Boosting Machine model explained 56% of the variance. The EBM model also showed lower prediction error, with RMSE = 7.67 and MAE = 5.94. Global importance analysis indicated that dissociation was the strongest predictor, followed by alexithymia and trauma exposure. Interaction analysis showed that the dissociation by alexithymia interaction was the most influential interaction effect.
Conclusion: The findings indicate that emotional numbness is best predicted by psychological processing variables, particularly dissociation and alexithymia, rather than trauma exposure alone. The Explainable Boosting Machine model provided both stronger predictive accuracy and clinically interpretable results, suggesting that emotional numbness may emerge through non-linear and interactive pathways involving trauma burden, dissociative detachment, and impaired emotional awareness.
