Intelligent Production and Detection Template of Outlier Dataset Using Clustering

Abstract

Outliers are data with anomalous behaviors to other datasets. There are three different types of outliers, namely point anomaly, collective anomaly, and conditional anomaly. Different density-, clustering-, distance-, and distribution-based methods are used to detect outliers. It is obvious that before testing detection algorithms, a dataset that encompasses different types of outliers is required. In this paper an intelligent clustering algorithm is presented to produce a dataset consisting of different outliers. The other important point in this paper is the probability of two uninvestigated types of collective data among datasets that the anomalies are called type I and II. Results show that the proposed algorithm is capable of producing a dataset including different types of outliers. This dataset can be used in all outlier detection techniques. In addition to detection of point anomalies, it can detect all collective anomalies.