AUTHOR=Mizrahi Dor , Zuckerman Inon , Laufer Ilan 

TITLE=Predicting attachment style from EEG data on the Flanker task

JOURNAL=Frontiers in Human Neuroscience

VOLUME=Volume 19 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/human-neuroscience/articles/10.3389/fnhum.2025.1651943

DOI=10.3389/fnhum.2025.1651943

ISSN=1662-5161

ABSTRACT=Bowlby's attachment theory describes the differences that people exhibit in the way they form emotional bonds with others. The dimensional measure of attachment describes it by the magnitude of anxiety and an avoidance dimension, which are currently measured using a self-report questionnaire. Recent advances in neurophysiological methods have started exploring the neural underpinnings of attachment styles. Nonetheless, a conspicuous gap remains: the underexplored realm of predictive models for predicting attachment styles based on objective physiological data. With that in mind, we have constructed a model for inferring individual attachment profiles, based solely on their brain signals recorded using an electroencephalogram (EEG). For that aim, we recorded EEG data of 27 participants engaged in the Flanker task and receiving either positive or negative feedback following each trial. We then utilized the recently developed ROCKET algorithm (RandOm Convolutional KErnel Transform) to automatically extract 20,000 time-series features from the EEG data. Next, we applied a Principal Component Analysis (PCA) and reduced the number of features to 87 individual components that were used to construct regression models predicting participants' anxiety and avoidance scores, as measured by the ECR-R questionnaire. Our results show, for the first time, that individual attachment profiles can be inferred from EEG data, allowing post hoc categorization into the four canonical attachment styles. This offers two key contributions: first, it provides an objective alternative to traditional self-report questionnaires, helping reduce subjectivity bias in attachment assessment. Second, it highlights the value of using automatically generated features over the limited set of hand-crafted features typically found in the literature.