Research On The Abnormal Trajectory Detection In Mobility Big Data

Mining abnormal trajectory from mobility big data is a key research in big data safety analysis research area. Comparing to the traditional abnormal trajectory detection research, this study is more difficulty due to the larger data volume and the complicated data catergory. One of the difficult problem is how to mine connection between moving objects from raw data and use this connection detecting abnormal trajectory of moving object, especially, the abnormal trajectory of irregular moving object. According to this problem, there are two research works have been done in this thesis:Firstly, A TF-IDF theory based similar user detection method(short as TTSUDM) has been proposed in this thesis. This method is use to solve the problem of mining connection between users from raw data and it can be divided into three parts: mining StayArea from mobility big data is the first step; then mining StayPoint from StayArea; using the theory of TF-IDF calculating the importance value of StayPoint and combining the cosine similarity calculating the similarity value among users is the final step. The experiment results show that the theory of this method is totally same with hypothesis proposed by this thesis and it can detect the similar users accurately.Secondly, an abnormal trajectory detection method among similar users(short as ATDM-SU) based on above research has been proposed in this thesis. This method is mainly use to solve the problem of detecting abnormal trajectory of irregular moving object and it can be divided into four parts: the first step is using TTSUDM detecting similar users and selecting the higher similarity value of user groups as detection objects; then mining trajectory of these users from raw data; the definition of trajectory neighbor points is proposed for calculating the connection between trajectories is the third step; finally, the definition of abnormal trajectory is proposed for detecting abnormal trajectory of irregular moving object. The experiment results show that this method has better performance and is more adapted to the environment of big data comparing to the traditional abnormal trajectory detection algorithm. The accuracy of this method is range from 50 percent to 70 percent in the special experiment environment of this thesis.