Comparative Analysis of Proximity Based Outlier Detection Algorithms

Abstract

Outlier detection is the process of finding peculiar pattern from given set of data. Nowadays, outlier detection is more popular subject in different knowledge domain. Data size is rapidly increases every year there is need to detect outlier in large dataset as early as possible. In this research, comparison of three different proximity based outlier detection algorithm i.e. distance based method (LDOF), cluster-based method (K-medoid based OD) and density based method (LOF) is presented. The main aim of this research is to evaluate their performance of those three different proximity based outlier algorithm for different dataset with different dimension. The dataset used for this research are chosen such way that they are different in size, mainly in terms of number of instances and attributes. When comparing the performance of all three proximity based outlier detection algorithms, density based method (LOF) is found to be better algorithm to detect outlier in most cases with accuracy level 94.47% as well as 66.93% precision, 83.14% recall and 73.18% F-measure value.