A Study On The Outlier Mining With A Practice In The Securities Industry

Within the studies of data mining, outlier mining has always been the hot issue in theory and practice. With the increasing concerns and research, outlier mining has experienced the evolution from the by-product of clustering to an independent topic with various algorithms and theories. Since people are realizing that outlier mining has been much more useful in defect detection and fraud avoiding, more and more cases of outlier mining are rising in banking, insurance, and telecommunication industry. Furthermore, with the fast developing in network and computing technology, outlier mining and other data mining technologies will have a brighter future.Looking back, most research on outlier mining is focused on the algorithms by means of mathematical statistics, set theory, discrete mathematics and computer science. But they are blind in how to apply outlier mining into the practical problem and how to be a dber. With the applications in fraud detection, it is necessary to answer the question.First of all, the paper will introduce the existing life-time model of data mining, the concepts of outlier and the algorithms for mining outliers. Some materials and papers will be quoted and some basic terms will be explained.Then, emphases are put on the life-time model of outlier mining, and try to answer the question: how to process a practical mining. 7 aspects and procedures are described and an example is raised to illustrate the theory.Meanwhile, 2 kinds of distance-based algorithms are implemented: DKP neighbor algorithm and LOF density algorithm both are vital for the rest of the paper.Finally, a real case will be raised under the direction of the theory described in former chapters. The case tries to prove the theory and use the algorithms to do a real mining, some reasonable advice will also be brought forward according to the mining outcomes. All these make the paper complete and logical in both theory and practice.