Abstract: Most outlier detection approaches suffer from parameter dependencies and parameter instability. In order to solve the two problems, this paper introduces a new notion of outlier based on reachable neighbor, which is called RN-Outlier, and proposes novel and fast algorithms for outlier detection. Experimental results of synthetic datasets and real datasets show that the algorithm outperforms both LOF and LSC methods. The algorithm is more stable when parameter changes.

Key words: outlier detection, reachable neighbor, local density, local outlier

摘要： 提出了可达邻域的概念，定义了基于可达邻域的异常RN-Outlier。给出了RNOF异常检测算法，克服了异常检测算法常被参数依赖和参数扰动所困扰的缺点。仿真数据集和真实数据集的实验表明，该算法的性能超过了经典的LOF和LSC算法，降低了参数依赖和参数扰动的影响。

关键词: 异常检测, 可达邻域, 局域密度, 局部异常

CLC Number: 

• TP393