Improving attachment style clustering with ROCKET and CatBoost: Insights from EEG analysis

Area of Science:

• Neuroscience
• Psychology
• Machine Learning

Background:

• Attachment styles are crucial in psychology and neuroscience.
• Predicting attachment styles using objective neural data is challenging.
• Existing methods lack objective neural markers for nuanced attachment classification.

Purpose of the Study:

• To explore the use of machine learning (ML) models and electroencephalography (EEG) analysis for improved attachment style classification.
• To investigate the relationship between EEG features and different attachment styles (secure, avoidant, anxious, fearful-avoidant).
• To assess the potential of ML-driven EEG analysis for psychological assessment.

Main Methods:

• EEG data were collected from 27 university students.
• Attachment styles were assessed using the ECR-R questionnaire.
• EEG features were extracted using the ROCKET algorithm, followed by Principal Component Analysis (PCA) and CatBoost for prediction, with a two-stage data pruning approach.

Main Results:

• A strong relationship was found between the number of EEG epochs and predictive accuracy.
• Secure and Fearful-Avoidant attachment styles were predicted most reliably.
• Anxious and Avoidant styles showed greater variability, indicating complex neural signatures.

Conclusions:

• Findings support attachment as a spectrum influenced by experiences, emotional regulation, and social context, rather than fixed categories.
• ML-driven EEG analysis shows potential for predicting attachment styles, offering new avenues for psychological assessment.
• The study highlights attachment as a dynamic process, informing clinical interventions and research on neural markers.