Awareness of Crime-related Information and Concealed Information Detection method

Concealed information test based on P300 and machine learning has become increasingly popular in the fields of cognitive psychology. Numerous studies have set up mock crime scenario to identify changes in EEG cognitive components. However, only two kinds of subjects are taken into account in most previous studies. Therefore, if an innocent person had been aware of case-related information, i.e., probe items, which is likely to happen in practice, recognition capability will be significantly compromised. In order to simulate practical cases, three kinds of subjects needed to be discriminated, including guilty, innocent and informed. 36 subjects went through a mock crime scenario, and EEG signals obtained on 8 electrodes were analyzed. After preprocessing, the discrete wavelet packet decomposition was used to extract EEG features. Subsequently, a multi-scale wavelet kernel extreme learning machine classifier is proposed to recognize the group to which a specific subject belongs. To further reduce computation, Cholesky decomposition is introduced during the calculation of the output weights. Our results demonstrate that the proposed algorithm can achieve good recognition performance and has low computational burden.