LieWaves: dataset for lie detection based on EEG signals and wavelets

Area of Science:

• Neuroscience
• Computer Science
• Forensic Science

Background:

• Lie detection is critical in legal, security, and professional settings.
• Existing methods for lie detection often lack objective physiological measures.
• Electroencephalography (EEG) offers a promising avenue for objective deception detection.

Purpose of the Study:

• To create a novel EEG dataset for lie detection research, named LieWaves.
• To analyze EEG signals for deception detection using advanced signal processing and deep learning.
• To establish a benchmark for future lie detection studies using EEG data.

Main Methods:

• Acquired 5-channel EEG data from 27 participants using Emotiv Insight during honest and deceitful trials.
• Applied Automatic and Tunable Artifact Removal (ATAR) for preprocessing and Overlapping Sliding Window (OSW) for data augmentation.
• Utilized a combination of Discrete Wavelet Transform (DWT), Fast Fourier Transform (FFT), and statistical methods for feature extraction.
• Classified features using Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and hybrid CNN-LSTM models.

Main Results:

• The novel LieWaves dataset was successfully created and preprocessed.
• Feature extraction combined DWT, FFT, and statistical methods.
• The Long Short-Term Memory (LSTM) model, in conjunction with Discrete Wavelet Transform (DWT), achieved a peak accuracy of 99.88% in lie detection.
• The study demonstrated the effectiveness of the developed dataset and methodologies.

Conclusions:

• The LieWaves dataset provides a valuable resource for advancing EEG-based lie detection research.
• Deep learning models, particularly LSTM combined with DWT, show high efficacy in identifying deception from EEG signals.
• This research addresses a gap in the literature by providing a comprehensive dataset and validated analysis techniques for lie detection.