Feature Optimize and Classification of EEG Signals: Application to Lie Detection Using KPCA and ELM

• Main Authors: GAO Junfeng, QIU Jianhui, ZHANG Wenjia, YANG Yong
• Format: Article
• Language: English
• Published: IFSA Publishing, S.L. 2014-04-01
• Series: Sensors & Transducers
• Subjects: Lie detection; Extreme learning machine; Kernel principal component analysis; P300.; EEG signal processing; Lie detection; Extreme learning machine; Kernel principal component analysis; P300.
• Online Access: http://www.sensorsportal.com/HTML/DIGEST/april_2014/Vol_169/P_RP_0109.pdf
• ISSN: 2306-8515; 1726-5479

EEG signals had been widely used to detect liars recent years. To overcome the shortcomings of current signals processing, kernel principal component analysis (KPCA) and extreme learning machine (ELM) was combined to detect liars. We recorded the EEG signals at Pz from 30 randomly divided guilty and innocent subjects. Each five Probe responses were averaged within subject and then extracted wavelet features. KPCA was employed to select feature subset with deduced dimensions based on initial wavelet features, which was fed into ELM. To date, there is no perfect solution for the number of its hidden nodes (NHN). We used grid searching algorithm to select simultaneously the optimal values of the dimension of feature subset and NHN based on cross- validation method. The best classification mode was decided with the optimal searching values. Experimental results show that for EEG signals from the experiment of lie detection, KPCA_ELM has higher classification accuracy with faster training speed than other widely-used classification modes, which is especially suitable for online EEG signals processing system.