An Algorithm for Computing Global-Based Outlier Degrees on Data Sets

Huge information can be easily collected now due to drastic improvement in computer throughput and cost reduction of mass memory medium. Recently, there is a strong expectation for establishment of a new technology called data mining for discovering useful knowledge from a lot of data in various fields. One of these technologies is t o detect outliers. Detecting outliers is a technology that is useful for detecting fraud, finding exceptional data and other by detecting abnormal data objects from data such as business data and network access log. The aim of this research is to assign a high outlier degree to a data object, which is globally separated from clusters. We propose a method of calculating global-based outlier degrees on data sets, and the proposed method is evaluated experimentally.