Depression recognition using machine learning methods with different feature generation strategies.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Current depression diagnosis relies on subjective communication and scales, leading to inaccuracies.
• Objective, automated methods are crucial for improving depression recognition and treatment accuracy.
• Electroencephalography (EEG) offers a potential objective biomarker for neurological and psychiatric conditions.

Purpose of the Study:

• To develop and evaluate machine learning methods for enhanced depression recognition using EEG features.
• To compare the performance of ensemble learning and deep learning approaches for EEG-based depression detection.

Main Methods:

• Recorded 128-channel EEG data from 28 subjects during an emotional face stimuli task.
• Extracted EEG features including power spectral density and activity using Auto-regressive and Hjorth algorithms.
• Applied ensemble learning (Deep Forest + SVM) and deep learning (CNN with image conversion) for classification.

Main Results:

• The ensemble model using power spectral density achieved the highest accuracy of 89.02%.
• The deep learning model using activity achieved an accuracy of 84.75%.
• Both methods demonstrated the efficiency of EEG in recognizing depression.

Conclusions:

• EEG features, when processed with machine learning, can reliably indicate depression.
• The proposed methods show promise for developing portable, auxiliary depression recognition systems.
• This research supports the future clinical application of EEG in mental health diagnostics.