On effective cognitive state classification using novel feature extraction strategies.

Area of Science:

• Neuroscience and Signal Processing
• Human-Computer Interaction

Background:

• Electroencephalography (EEG) signal analysis is crucial for understanding human cognitive states.
• Existing feature extraction methods for ambulatory EEG have limitations in classification accuracy.

Purpose of the Study:

• To develop a cost-effective system for cognitive state classification using ambulatory EEG signals.
• To introduce and evaluate a novel feature extraction method, Gammatone Cepstrum Coefficients (GTCC), for EEG analysis.

Main Methods:

• A novel event-driven environment with external stimuli was used to capture 14-channel Emotiv neuro-headset EEG data.
• Gammatone Cepstrum Coefficients (GTCC) were introduced and compared against Discrete Wavelet Transformation (DWT) and Mel-Frequency Cepstral Coefficients (MFCC) using Fisher Discriminant Ratio (FDR) and Logistic Regression (LR).
• Ensemble feature spaces (GTCC+MFCC) and features extracted via a 1D Convolutional Neural Network (CNN) from ensemble sets were evaluated with various classifiers including Deep Convolutional Generative Adversarial Network (DCGAN).

Main Results:

• GTCC features demonstrated superior discriminative power compared to DWT and MFCC.
• An ensemble feature space of GTCC and MFCC, along with GTCC+MFCC+CNN features, significantly improved classification performance.
• The proposed system achieved high accuracies of 96.42% (GTCC+MFCC) and 96.14% (GTCC+MFCC+CNN) with the DCGAN classifier, outperforming state-of-the-art methods.

Conclusions:

• The novel GTCC feature extraction method offers significant advantages for ambulatory EEG signal analysis.
• Ensemble feature spaces and CNN-extracted features provide superior classification accuracy for cognitive states.
• The developed system demonstrates high efficacy and potential for cognitive science applications.