New Method of Reverse Neighbour Counts in Unsupervised Outlier Detection

Abstract

: Outlier detection is the process of finding outlying pattern from a given dataset. Outlier detection became important subject in different knowledge domains. Data size is getting doubled every years there is a need to detect outliers in large datasets as early as possible. In high-dimensional data outlier detection presents various challenges because of curse of dimensionality. By examining again the notion of reverse nearest neighbors in the unsupervised outlier-detection context, high dimensionality can have a different impact. In high dimensions it was observed that the distribution of points in reverse-neighbor counts becomes skewed .This proposed work aims at developing and comparing some of the unsupervised outlier detection methods and propose a way to improve them. This proposed work goes in details about the development and analysis of outlier detection algorithms such as Local Outlier Factor(LOF), Local Distance-Based Outlier Factor(LDOF) , Influenced Outliers and .The concepts of these methods are then combined to implement a new method with distributed approach which improves the results of the previous mentioned ones with reference to speed, complexity and accuracy.