KMAN Publication Inc. (KMANPUB) Journal of Assessment and Research in Applied Counseling (JARAC) 8 Serial Number 28 2026 01 01 Modeling Social Anxiety, Attentional Bias, and Metacognitive Beliefs with Supervised Learning Algorithms Modeling Social Anxiety, Attentional Bias, and Metacognitive Beliefs with Supervised Learning Algorithms 1 11 10.61838/kman.jarac.5181 EN Journal Article 2025 09 17 Objective: The objective of this study was to utilize supervised machine learning algorithms to model and predict the severity of social anxiety based on the complex, non-linear interplay between multidimensional metacognitive beliefs and objective chronometric measures of attentional bias. Methods and Materials: A cross-sectional, predictive study design was employed with 854adult participants (mean age 34.52years) recruited from the USA via online crowdsourcing. Social anxiety was assessed using the Liebowitz Social Anxiety Scale (LSAS), and metacognitive beliefs were measured via the Metacognitions Questionnaire-30 (MCQ-30). Attentional bias was quantified using a computerized visual dot-probe paradigm, measuring reaction times in milliseconds. Following preprocessing and k-nearest neighbors imputation for missing values (approximately 2%), the data were split into an 80%training and 20%testing set. Supervised learning algorithms, including Random Forest, Support Vector Machines, and Gradient Boosting Regressors, were trained using 10-fold cross-validation. Model evaluation relied on R^2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE), while feature importance was interpreted using SHapley Additive exPlanations (SHAP) values. Findings: On the unseen testing set (n=171), the Gradient Boosting Regressor emerged as the optimal model, demonstrating superior predictive performance (R^2=.68, RMSE=13.65, MAE=10.42) compared to the Random Forest Regressor (R^2=.63, RMSE=14.68, MAE=11.55) and the Support Vector Machine (R^2=.55, RMSE=16.18, MAE=12.89). Feature importance analysis using SHAP values indicated that Negative Beliefs about uncontrollability and danger had the highest predictive influence (Mean ∣SHAP∣=6.84), closely followed by the Attentional Bias Score (Mean ∣SHAP∣=4.72) and the Need to Control thoughts (Mean ∣SHAP∣=3.55). Furthermore, SHAP dependence plots revealed a distinct non-linear interaction wherein the predictive impact of attentional bias was exponentially magnified in the presence of high negative metacognitive beliefs. Conclusion: Advanced machine learning algorithms successfully captured the complex architecture of social anxiety, demonstrating that clinical severity is predominantly driven by a synergistic, non-linear interaction between severe negative metacognitions and bottom-up visual hyper-vigilance toward threat.

https://journals.kmanpub.com/index.php/jarac/article/download/5181/9414