Campus Crowd Pattern Mining And Anomaly Detection Based On Wi-Fi Data

At end of the last century, Wi-Fi technology has been started up, and its continuously development leads to the low deployment costs and high performance. At present, Wi-Fi has become the most preferred way to get online in a fixed place. As a result, massive data has been generated by Wi-Fi network, which makes Wi-Fi based data mining a popular research field. Because of the intrinsic "time and spatial" attribute of Wi-Fi data,its data mining result can illustrate people's behavior patterns more thoroughly, making the research results and applications of this field have a great impact on people's life.This paper proposed two new methods for crowd pattern mining and population anomaly detection based on Wi-Fi data, which is applicable to campus scene. In the first method, the density-based crowd pattern mining algorithm is proposed. First, a two-level SVM classifier is used to detect users behavior category, which are summed up to obtain crowd patterns. Then, we use statistical algorithm, based on mean and entropy calculation, to detect abnormal changes in the crowd. In the second method, the topic-based crowd pattern mining algorithm is proposed. The algorithm transforms Wi-Fi data into series "word-analogues" data by clustering and generates the document data. Then, based on topic model the unsupervised mining of crowd pattern is accomplished. Finally, using difference between optimal estimate obtained by Kalman-Filter and actual observations, crowd anomaly can be detected. The experiment results based on the data of Beijing University of Posts and Telecommunications campus Wi-Fi shows that, the algorithms proposed in this paper can not only find out potential population distribution patterns on campus, but also effectively detect all kinds of campus crowd abnormalities.