Psychiatry Investig.  2025 Apr;22(4):412-423. 10.30773/pi.2024.0392.

Machine Learning Prediction of Attachment Type From Bio-Psychological Factors in Patients With Depression

• Cho YJ
• Ryu JS
• Seok JH
• Kim E
• Oh J
• Kim BH

• KMID: 2566735
• DOI: http://doi.org/10.30773/pi.2024.0392

Abstract

Objective
Adult attachment style is linked to how an individual responds to threats or stress and is known to be related to the onset of psychiatric symptoms such as depression. However, as the current assessment of attachment type mainly relies on self-report questionnaires and can be prone to bias, there is a need to incorporate physiological factors along with psychological symptoms and history in this process. We aimed to predict the measurement of two important types of adult attachment with heart rate variability (HRV), early life stress experience, and subjective psychiatric symptoms.
Methods
Five hundred eighty-two subjects with depressive disorder were recruited retrospectively from January 2015 to June 2021. The experience of early life stress and psychiatric symptoms were collected, and HRV measures were obtained as input for an ensembled Voting Regressor model of machine learning-based regression models, including linear regression, ElasticNet, Support Vector Machine (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost).
Results
Model performances evaluated with R-squared score averaged across 30 seeds were 0.377 and 0.188 for anxious- and avoidant-attachment, respectively. Mean absolute error averaged to 13.251 and 12.083, respectively. Shapley value importance analysis indicated that for both attachment types, the most important feature was the trait-anxiety, followed by emotional abuse, state-anxiety or self-reported depressive symptoms, and fear or helplessness felt in the moment of an early life stressor.
Conclusion
Our results provide the evidence base that may be utilized in clinical settings to predict the degree of attachment type using bio-psychological factors.

Keyword

Object attachment; Machine learning; Depressive disorder; Early life stress; Autonomic nervous system