Classification of alcoholic EEG signals using wavelet scattering transform-based features.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Feature extraction is crucial for machine learning and data science.
• The wavelet scattering transform (WST) is a novel technique preserving high-frequency information and signal deformation insensitivity.
• WST generates low-variance, real-valued features suitable for classification.

Purpose of the Study:

• To evaluate the efficacy of WST-based features for discriminating between alcoholic and healthy subjects using EEG signals.
• To compare WST performance with Convolutional Neural Networks (CNNs) and conventional classifiers.
• To identify informative EEG regions for alcoholism detection.

Main Methods:

• Utilized multichannel electroencephalogram (EEG) data from a public UCI database.
• Extracted features using the wavelet scattering transform (WST).
• Employed 10-fold cross-validation with Support Vector Machine (SVM), 1D CNN, and Linear Discriminant Analysis (LDA) classifiers.

Main Results:

• WST features with SVM achieved perfect classification of alcoholic and normal EEG records.
• 1D CNN demonstrated similar high performance.
• WST features combined with LDA yielded the highest independent-subject-wise cross-validation performance.
• Occipital and parietal EEG regions provided the most discriminative features.

Conclusions:

• The wavelet scattering transform (WST) offers a viable alternative to CNNs for classifying alcoholic and normal EEGs.
• WST combined with conventional classifiers provides a powerful tool for EEG-based alcoholism detection.
• Specific EEG regions, particularly occipital and parietal, are key indicators for differentiating alcoholism.