Comparative analysis of ROCKET-driven and classic EEG features in predicting attachment styles

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychology

Background:

• Predicting attachment styles using AI is an emerging research area.
• Attachment styles influence interpersonal relationships and mental well-being.
• Electroencephalography (EEG) offers a neurophysiological measure for psychological states.

Purpose of the Study:

• To compare the effectiveness of ROCKET-driven features versus classic features for classifying attachment styles using EEG data.
• To evaluate the performance of the XGBoost machine learning algorithm in attachment style prediction.
• To address the gap in scientific literature regarding AI-driven attachment style prediction.

Main Methods:

• Participants completed the ECR-R questionnaire to assess attachment styles.
• EEG data was collected during the Arrow Flanker Task with feedback.
• XGBoost algorithm analyzed ROCKET-derived and classic features for classification.

Main Results:

• Both feature sets showed effectiveness in classifying attachment styles.
• ROCKET-derived features achieved an 88.41% True Positive Rate (TPR) for insecure attachment.
• ROCKET-derived features demonstrated superior performance across multiple metrics compared to classic features.

Conclusions:

• AI, particularly with ROCKET-derived features, shows significant potential for psychological assessments.
• Feature selection is crucial for optimizing AI model performance in specific applications.
• This study advances the integration of EEG and machine learning for understanding attachment styles.