Ensemble Wavelet Decomposition-Based Detection of Mental States Using Electroencephalography Signals

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Automation in industries necessitates intelligent machinery for Brain-Computer Interface (BCI) systems.
• Electroencephalography (EEG) offers a low-cost, non-invasive method for BCI, but its complex signal nature hinders manual analysis.
• Accurate, automatic mental state detection is crucial for advancing BCI technology.

Purpose of the Study:

• To develop an automated system for classifying human mental states (focused, unfocused, drowsy) using EEG signals.
• To investigate the efficacy of an ensemble of wavelet transforms for EEG signal decomposition and feature extraction.
• To optimize an ensemble classifier for improved mental state detection accuracy in BCI applications.

Main Methods:

• Employed an ensemble of tunable Q wavelet transform, multilevel discrete wavelet transform, and flexible analytic wavelet transform for EEG signal processing.
• Extracted features from subbands of EEG signals corresponding to focused, unfocused, and drowsy mental states.
• Utilized an optimized ensemble classifier, incorporating feature fusion and iterative majority voting, for classification.

Main Results:

• Feature fusion from ensemble decomposition led to dimensionality reduction.
• The proposed model achieved high classification accuracies of 92.45% (ten-fold cross-validation) and 97.8% (iterative majority voting).
• Demonstrated the effectiveness of the combined wavelet transforms and ensemble classification approach.

Conclusions:

• The developed automated system accurately detects mental states from EEG signals.
• The proposed method, leveraging wavelet transform ensembles and optimized classification, significantly enhances BCI system performance.
• This approach is suitable for real-time mental state detection, paving the way for more sophisticated BCI applications.