Lie detection has been widely used by governmental and non-governmental organizations to ensure the reliability of criminal confessions. Conventional polygraph devices, however, suffer from limitations and inconsistent accuracy. This paper presents a novel approach for lie detection using electroencephalogram (EEG) signals. An EEG dataset was collected from 20 participants, and a six-layer graph convolutional network (GCN) integrated with type-2 fuzzy sets was employed for feature selection and automatic classification. Experimental results demonstrate that the proposed method achieves over 90% accuracy even in noisy environments (SNR = 0 dB), outperforming existing techniques and showing strong potential for practical applications.
