Extreme anomalous clustering algorithm

Abstract

A clustering algorithm is a process of dividing data points into disjoint clusters according to the similarity of the data points. Many researchers have presented and developed the clustering algorithms to be suitable for the different characteristics of the datasets. Therefore, the concept of the identifying anomaly data points in the dataset was adopted in this thesis to explain the similarity of the data points. The main idea is to calculate the minimum distance between data points to identify the extreme anomalous score of all data points. In this thesis, a novel clustering algorithm is proposed called the Extreme Anomalous Clustering algorithm or EAC. This algorithm specifies the extreme anomalous scores for all data points in the dataset and combines two data points with the smallest extreme anomalous score. Then the algorithm selects one data point within the cluster as the representative point which is used to consider the next combination. The experiments of this thesis are created to compare the performance of the EAC algorithm on the simulated datasets and UCI datasets with AGNES, k-means, and DBSCAN. The experimental results show that the EAC algorithm is better than the three algorithms according to the Silhouette and Rand index measurements.