Fighting Terrorism: Semi-Supervised Outlier Detection of Electrical Power Consumption

Among years of experiences in counter-terrorism investigative practices, it has been found that terrorist organizations and their activities typically exhibit an exceptional demand for electrical power. For example, making tools for criminal purpose such as knives and bombs, or performing underground preaching, concealed gatherings and other illegal activities, might contribute to unusual fluctuation of power consumption. Intriguingly, facilitated by monitoring key knowledgeable personnel behavior, learning anomalies from massive electrical power consumption data helps to timely obtain predictive clues and warnings for fighting terrorism. Along this line, in this paper we propose a semi-supervised approach incorporating power consumption and key personnel monitoring to draw insights on counter-terrorism. Specifically, we: (1) extract features differentiating normal and abnormal power consumption patterns; (2) construct a semi-supervised method, including unsupervised stage to discover suspicious exceptional power consumption from massive data, and supervised stage leveraging key knowledgeable personnel behavior to refine targets; and (3) deploy a prototype system in Xinjiang China, collaborated with local authorities, which practically testifies the interesting association between power consumption and counterterrorism. Besides, experiments demonstrate satisfactory performance of proposed approach. Indeed, the innovation of combining power consumption and key knowledgeable personnel behavior in perceiving terrorism in advance has profound theoretical and practical significance.