TensorCSBP: A Tensor Center-Symmetric Feature Extractor for EEG Odor Detection

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Accurate odor classification from electroencephalography (EEG) signals is crucial but limited by feature explainability and tensor-level implementations.
• Existing methods like Local Binary Pattern (LBP) lack sufficient interpretability for neuroscientific insights into odor processing.

Purpose of the Study:

• To introduce Tensor Center-Symmetric Binary Pattern (TensorCSBP) as a novel tensor-based feature extractor for EEG odor analysis.
• To develop an explainable feature engineering (XFE) pipeline for enhanced EEG-based odor classification.

Main Methods:

• The proposed TensorCSBP method was integrated into a four-step XFE pipeline: TensorCSBP feature generation, CWNCA feature selection, tkNN classification, and DLob symbolic interpretability.
• The pipeline was evaluated on a new 32-channel EEG dataset specifically collected for odor detection tasks.

Main Results:

• The TensorCSBP XFE pipeline achieved a high accuracy of 96.68% on the odor detection task, validated through 10-fold cross-validation.
• The information entropy of the DLob symbol sequence was calculated to be 3.5675, indicating a rich interpretability output.

Conclusions:

• This study presents a highly accurate, explainable, and computationally efficient model for EEG-based odor classification.
• TensorCSBP effectively bridges low-level signal patterns with symbolic neuroscience insights, demonstrating potential for real-time Brain-Computer Interface (BCI) and clinical applications.