Performance Evaluation of Density-Based Outlier Detection on High Dimensional Data

Outlier detection is a task that finds objects that are considerably dissimilar, exceptional or inconsistent with respect to the remaining data. Outlier detection has wide applications which include data analysis, financial fraud detection, network intrusion detection and clinical diagnosis of diseases. In data analysis applications, outliers are often considered as error or noise and are removed once detected. Approaches to detect and remove outliers have been studied by several researchers. Density based approaches have been proved to be effective in detecting outliers successfully, but usually requires huge amount of computations. In this paper, two approaches that enhance the traditional density based method for removing outliers are analyzed. The first method uses data partitioning method and usespeed up strategies to avoid large computations. The second method presents a unified clustering and outlier detection using Neighbourhood based Local Density Factor (NLDF). The aim of both the models is to improve the performance of outlier detection, clustering and to speed up the whole process. In this paper, the working of these two papers is studied and a performance evaluation based on clustering efficiency and outlier detection efficiency is presented.