A novel classification method based on ICA and ELM: a case study in lie detection

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Accurate lie detection remains a significant challenge in forensic science and psychology.
• Electroencephalography (EEG) offers a non-invasive method to study brain activity related to deception.
• Existing methods for analyzing EEG signals in lie detection often require complex feature engineering and extensive training.

Purpose of the Study:

• To propose a novel classification model for lie detection using EEG signals.
• To enhance the accuracy and efficiency of identifying deceptive individuals.
• To compare the performance of the proposed model against traditional machine learning classifiers.

Main Methods:

• Independent Component Analysis (ICA) was employed to automatically identify P300 independent components (ICs) from EEG data.
• Time and frequency-domain features were extracted from reconstructed P3 waveforms.
• Extreme Learning Machine (ELM), Back-propagation Network (BPNN), and Support Vector Machine (SVM) classifiers were trained and compared.
• Cross-validation was used to optimize classifier parameters and the number of P3 ICs.

Main Results:

• The proposed ICA-ELM model achieved a high training accuracy of 95.40% in detecting P3 components.
• The ICA-ELM method demonstrated significantly reduced training and testing times compared to other classifiers.
• The model effectively distinguished between guilty and innocent subjects based on EEG P3 components.

Conclusions:

• The developed ICA-ELM model shows high efficacy and efficiency for lie detection using EEG.
• This method offers a promising advancement in the field of neuro-forensic science.
• The approach has the potential for practical application in real-world lie detection scenarios.