利用MapReduce的异常轨迹检测并行算法

异常轨迹检测是移动对象数据挖掘的一个重要研究领域。TRAOD(TRAjectory Outlier Dectection Algorithm)算法是一种经典的异常轨迹检测算法,但它对于海量轨迹数据的异常检测效率低。为提高海量轨迹数据集的异常检测效率,本文提出了一种利用MapReduce 的异常轨迹检测并行算法(Parallel algorithm for TRAjectory Outlier Detection, PTRAOD),并在此基础上提出了网格索引的异常轨迹检测并行算法(Grid-based Parallel algorithmfor TRAjectory Outlier Dectection, GPTRAOD)。GPTRAOD算法在PTRAOD算法的基础上,利用网格索引实现区域查询,进一步提高算法效率。将PTRAOD算法和GPTRAOD算法在Hadoop 平台上加以实现,结果表明:本文提出的2 个并行检测算法,能实现异常轨迹的检测;GPTRAOD算法的效率优于PTRAOD算法;GPTRAOD算法具有较高的可扩展性和较好的加速比。

A Parallel Algorithm for Detecting Trajectory Outliers Based on MapReduce

Trajectory outlier detection is significantly important in the field of data mining for moving object. TRAOD (TRAjectory Outlier Dectection Algorithm), a classic algorithm for detecting trajectory outliers, focuses on a new two-level trajectory partitioning strategy to enhance the efficiency of algorithm. The main advantage of TRAOD algorithm is the ability to detect outlying sub-trajectories. However, it has a low efficiency on abnormality detection for massive trajectory data. In order to improve the efficiency for mining trajectory outliers from massive datasets, a parallel algorithm for detecting trajectory outliers based on MapReduce framework, which is called PTRAOD (Parallel algorithm for TRAjectory Outlier Detection), is presented. It redesigns the TRAOD algorithm based on the MapReduce framework, and encapsulates the steps of TRAOD into its Map and Reduce functions. PTRAOD algorithm takes full advantages of the features from Hadoop platform. It firstly distributes the trajectory data into distributed computing nodes. While distributing the data, it also takes the load-balance into consideration. And after all, each node runs the same algorithms to detect abnormal trajectories. Based on PTRAOD algorithm, a grid-based parallel algorithm for detecting trajectory outliers, called GPTRAOD (Gridbased Parallel algorithm for TRAjectory Outlier Detection), is then proposed. GPTRAOD algorithm makes use of the grid index to realize regional query and reduce unnecessary calculations. At first, GPTRAOD algorithm divides the map into a series of equal- sized grids, whose size is determined with respect to each specific data. Then, the grid index is established to implement the regional query. Finally, the algorithm finds out the abnormal trajectory segments and judges whether the trajectories that contains the abnormal trajectory segments are abnormal. In general, GPTRAOD algorithm takes advantages of the gird index to realize regional query on the basis of PTRAOD algorithm, which furthermore can search abnormal trajectory on the cloud computing platform. To assess the performances of the proposed algorithms, extensive experiments were conducted. The experimental results demonstrate that the proposed two parallel detection algorithms can both successfully achieve the trajectory outlier detection. The efficiency of PTRAOD algorithm is higher than TRAOD algorithm, while GPTRAOD algorithm has the higher scalability and better speedup ratio than PTRAOD algorithm. In addition, with the rapidly expanding of datasets, GPTRAOD algorithm shows obvious advantages and increasing potentials.