HOT Hypergraph#173;based Outlier Test for Categorical Data.pdfVIP

HOT: Hypergraph-based Outlier Test for Categorical Data Aoying Zhou, Li Wei, Weining Qian Wen Jin Dept. of Computer Science, Fudan Univ. Dept. of Computer Science, Simon Fraser Univ. ayzhou, lwei, wnqian@ wjin@cs.sfu.ca ABSTRACT Outlier detection is one of the major technologies in data As a widely used data mining technique, outlier detection is mining, whose task is to find small groups of data objects a process which aims to find anomalies while providing good that are exceptional when compared with rest large amount explanations. Most existing detection methods are basically of data. Outlier mining has strong application background designed for numeric data, however, real-life data such as in telecommunication, financial fraud detection, and data web pages, business transactions and bioinformatics records cleaning, since the patterns lying behind the outliers are always contain categorical data. So it causes difficulty to usually interesting for helping the decision makers to make find reasonable exceptions in the real world applications. profit or improve the service quality. In this paper, we introduce a novel outlier mining method A descriptive definition of outliers is given by Hawkins like based on hypergraph model for categorical data. Since hy- this:” an outlier is an observation that deviates so much from pergraphs precisely capture the distribution characteristics other observations as to arouse suspicions that it was gener- in data subspaces, this method is effective