Modeling Generalized Anxiety Disorder from Intolerance of Uncertainty and Attentional Control Using AI‑Driven Feature Selection

Abstract

Objective: The primary objective of this study was to identify the most robust, item-level cognitive predictors of generalized anxiety disorder severity utilizing an artificial intelligence-driven hybrid feature selection pipeline applied to the constructs of intolerance of uncertainty and attentional control. Methods and Materials: A cross-sectional predictive study design was utilized with a sample of 486 Canadian adults. Participants completed standardized self-report measures, including the 7-item Generalized Anxiety Disorder Assessment (GAD-7), the 12-item Intolerance of Uncertainty Scale (IUS-12), and the 20-item Attentional Control Scale (ACS). Data preprocessing involved k-nearest neighbors imputation for missing values and standardization of continuous variables (z=(x-μ)/σ). A hybrid machine learning approach combining Least Absolute Shrinkage and Selection Operator (LASSO) regression and Extreme Gradient Boosting Recursive Feature Elimination (XGBoost-RFE) was employed to extract key item-level predictors. The dataset was split into an 80%training set and a 20%testing set, with model hyperparameter tuning validated via 10-fold cross-validation. Findings: The sample (N=486; 61.3%female; age M=34.2, SD=10.5) exhibited significant baseline correlations between GAD-7 and both IUS-12 (r=.64,p<.001) and ACS (r=-.52,p<.001). From an initial pool of 32cognitive items, the hybrid AI pipeline (using an optimal LASSO penalty of α=0.034) identified a critical subset of just 7key features (4 capturing distress/paralysis from uncertainty and 3capturing difficulties in attentional shifting/focusing). The optimized XGBoost model utilizing these 7selected features achieved superior predictive performance on the test set (R^2=0.73, RMSE=2.40) compared to the baseline model utilizing all 32features (R^2=0.61, RMSE=2.88). Conclusion: Granular, item-level cognitive vulnerabilities—specifically uncertainty-induced paralysis and severe deficits in attentional shifting—are the primary drivers of anxiety severity, demonstrating the transformative potential of AI in precise psychiatric modeling.