[论文]郑宇军等人.Evolutionary ensemble generative adversarial learning for identifying terrorists among high-speed rail passengers
Evolutionary ensemble generative adversarial learning for identifying terrorists among high-speed rail passengers
August 2022 Expert Systems with Applications 210(196):118430
Authors:Yujun Zheng,Cong-Cong Gao,Yu-Jiao Huang;Wei-Guo Sheng,Zidong Wang
As one of the most salient features of China’s economic development, high-speed rail (HSR) is considered to be an attractive target and travel mode for terrorists. Distinguishing potential terrorists from normal passengers is of critical importance to public security, but very challenging because terrorists constitute only a very small fraction of HSR passengers, especially when they can disguise their attributes and behaviors to deceive the classifiers. For this extremely imbalanced classification problem, we propose a novel evolutionary generative adversarial network (GAN) ensemble method, where each GAN in the ensemble simultaneously trains a discriminator to identify abnormal samples from a large number of passenger profiles and trains a generator to produce abnormal samples that are disguised as normal ones in a subspace of the sample space, and the final classifier combines these GANs using an evolutionary fusion method. Experiments on benchmark problems demonstrate that the proposed method has very competitive performance compared to popular imbalanced classifiers. The successful applications in terrorist identification for China Railway also demonstrate the feasibility and effectiveness of our approach.