A Novel Multi-Scale Entropy Approach for EEG-Based Lie Detection with Channel Selection.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) offers objective lie detection potential, surpassing traditional polygraph methods.
• Entropy-based analyses quantify biological signal complexity and information content.
• Existing methods lack multidimensional characterization of EEG signals for deception detection.

Purpose of the Study:

• To develop a novel multi-scale entropy approach for enhanced EEG-based lie detection.
• To identify representative brain regions and channels for deception detection.
• To evaluate the generalizability of the proposed lie detection method.

Main Methods:

• A fused multi-scale entropy approach combining fuzzy entropy (FE), time-shifted multi-scale fuzzy entropy (TSMFE), and hierarchical multi-band fuzzy entropy (HMFE).
• Application of machine learning classifiers (LDA, SVM) on fused feature vectors for lie detection.
• Subject-dependent and cross-subject experiments using the LieWaves dataset, including channel selection analysis.

Main Results:

• Subject-dependent experiments achieved accuracies up to 82.74% (LDA, LOOCV).
• Cross-subject experiment yielded 64.07% accuracy (RBF-SVM, LOSOCV).
• Parietal midline (PZ) and left temporal (T7) channels were identified as most representative for lie detection.

Conclusions:

• The proposed multi-scale entropy fusion method effectively characterizes EEG signals for lie detection.
• PZ and T7 channels are crucial for identifying neural signatures of deception.
• Findings support the development of portable, fewer-channel EEG lie detection devices and offer insights into neural dynamics of lying.