基于方形邻域和裁剪因子的离群点检测方法

Square Neighborhood and Pruning Factor Based Outlier Detection Algorithm

针对改进的局部稀疏系数(Enhanced Local Sparsity Coefficient,简称ELSC)算法在邻域查询过程中存在的不足,以及为了提高算法查准率,提出了一种基于方形邻域和裁剪因子的离群点检测算法.首先采用方形邻域,吸取网格算法的思想,以扩张的方形邻域代替网格分割,快速地排除聚类点,避免了网格算法的"维灾"问题.其次为了提高算法的精确度,引入裁剪因子的概念对候选离群点集进行精选.最后通过新定义的局部稀疏指数确定离群点.试验测试表明,该算法的执行效率与检测精度均优于ELSC算法.

In viewof the shortcomings of enhanced local sparsity coefficient( Enhanced Local Sparsity Coefficient,called ELSC) algorithm in the process of neighborhood query,to improve the accuracy of the algorithm,this paper proposed a algorithm which square neighborhood and pruning factor based outlier detection algorithm. First of all,algorithm applied to the square neighborhood,which absorbs the idea of grid based algorithm,eliminates the normal points with dense square neighborhood rapidly;the algorithm partitioned dataset with square neighborhood,not with spatial girds,and overcomes the"dimension disaster"based on grid algorithm. Secondly,the identify accuracy could be improved within the novel pruning factor,which is used for the selection of candidate outlier points. In the end,the newly defined local sparse index could determine outlier data objects. Experimental result shows the algorithm is not only efficient in detection accuracy but also more effective than ELSC in the computation.