Outlier detection by active learning

摘要: Most existing approaches to outlier detection are based on density estimation methods. There two notable issues with these methods: one is the lack of explanation for flagging decisions, and other relatively high computational requirement. In this paper, we present a novel approach classification, in an attempt address both issues. Our isbased key ideas. First, simple reduction via procedure that involves applying classification labeled data set containing artificially generated examples play role potential outliers. Once task has been reduced then invoke selective sampling mechanism active learning problem. We empirically evaluate proposed using number sets, find our method superior methods same but standard also show it competitive state-of-the-art literature estimation, while significantly improving complexity explanatory power.