基于轨迹点局部异常度的异常点检测算法

Trajectory Outliers Detection Based on Local Outlying Degree

随着大量的定位数据被收集在应用服务器,如何从大量定位轨迹数据挖掘异常信息已逐渐成为一个令人关注的研究课题.针对当前流行的、以轨迹片段表示局部特征的异常点检测算法存在的问题,文中提出了以轨迹点表示局部特征的异常点检测算法TraLOD.该算法不仅提出了将每个轨迹点赋予一个0~1的值来表示其局部异常程度,而且还引入了相对距离来计算轨迹片段之间的不匹配性.此外,针对数据挖掘算法效率低的缺点,TraLOD引入了R-Tree和距离特征矩阵来提高算法效率.性能分析和实验都证明了TraLOD的有效性.

Along with more and more location data being collected in application servers,how to mine outliers from these trajectory datasets is becoming an interesting topic.Aiming to the problems that were brought by using segment as local feature in trajectory outlier detection,the paper not only presents the concept of local outlying degree to express the outlying degree of local feature,but also introduces relative distance to compute the dismath between two segments.Moreover,to fast the performance,R-Tree and distance feature matrix are introduced.Finally,both of detailed experiments and performance analysis prove the algorithm available.