A Novel Density-Based Outlier Detection Approach for Low Density Datasets

Abstract

Outlier detection has been seen as one of the important technique in data mining and analysis, which can discover anomalous behaviors of objects in a dataset. Although it has been successfully used in many domains (network intrusion detection, credit card fraud detection, medical diagnosis, etc.), its performance is not good for low density datasets, wherein the density of the outlier is similar to the density of its neighbors. In this paper, we aim to address the outlier detection problem for low density dataset. To this end, we design a novel relative local density-based outlier factor (RLDOF) to measure the outlier-ness of objects, based on which the densities of an object and its neighbors are redefined and calculated in a different way compared to existing approaches. The performance of RLDOF is evaluated on a set of artificial and real world datasets. The experimental results show that RLDOF could effectively improve the performance of outlier detection compared to existing approaches.