Clustering anxiety and depression among international student athletes during study abroad using transformer-based embeddings
Open Access
- 30 October 2025
- journal article
- Published by Sergii Iermakov in Physical Education of Students
- Vol. 29 (5) , 383-394
Abstract
Background and Study Aim. Student-athletes studying abroad experience increased risks of anxiety and depression as they balance academic and athletic responsibilities in cross-cultural environments. These psychological challenges may vary depending on individual adaptation, motivation, and environmental stressors. Although previous studies have applied network analysis to explore the structure of anxiety and depression symptoms, the relative effectiveness of advanced semantic approaches in identifying subgroup heterogeneity remains of practical interest. As a follow-up study, this research aimed to apply a transformer-based semantic embedding approach to cluster the mental health profiles of international student-athletes and to compare the model’s performance with traditional clustering methods. Materials and Methods. Data were collected from 219 Chinese international student-athletes who completed the GAD-7 and PHQ-9 questionnaires. Three models were compared: (1) K-means clustering on raw item scores, (2) K-means clustering after dimensionality reduction using Principal Component Analysis (PCA), and (3) K-means clustering on pseudo-text representations embedded via a transformer model, followed by PCA and K-means. Internal validity was assessed with silhouette scores. Between-cluster differences were analyzed using t-tests with Holm correction, effect sizes (Cohen’s d), and cluster profiles. Results. Model 3 (Transformer embeddings + PCA + K-means) outperformed Models 1 and 2, achieving the highest silhouette score (0.391). Visualization in 2D and 3D projections confirmed clearer separation. Three clusters were identified: Cluster 2 (high symptoms), Cluster 0 (intermediate), and Cluster 1 (low symptoms). Pairwise comparisons revealed significant differences across nearly all items. Conclusions. Transformer-based semantic embeddings provide an effective approach to clustering psychological symptoms, outperforming traditional numerical methods. The results indicate the heterogeneity of anxiety and depression subgroups among student-athletes during study abroad, offering valuable insights for targeted screening, early identification, and long-term monitoring.Keywords
This publication has 48 references indexed in Scilit:
- Big data analysis of mental health intervention effects in student-athletes: based on data mining techniques and affective computing algorithmsMultimedia Tools and Applications, 2024
- Transformer Autoencoder for K-means Efficient clusteringEngineering Applications of Artificial Intelligence, 2024
- EHR-BERT: A BERT-based model for effective anomaly detection in electronic health recordsJournal of Biomedical Informatics, 2024
- Using machine learning to forecast symptom changes among subclinical depression patients receiving stepped care or usual careJournal of Affective Disorders, 2023
- Cultural values, shame and guilt, and expressive suppression as predictors of depressionInternational Journal of Intercultural Relations, 2022
- Stratified Care vs Stepped Care for DepressionJAMA Psychiatry, 2021
- Can network analysis transform psychopathology?Behaviour Research and Therapy, 2016
- The prevalence and correlates of depression, anxiety, and stress in a sample of college studentsJournal of Affective Disorders, 2015
- A Brief Measure for Assessing Generalized Anxiety DisorderArchives of internal medicine (1960), 2006
- Silhouettes: A graphical aid to the interpretation and validation of cluster analysisJournal of Computational and Applied Mathematics, 1987