Performance Analysis of Rule Based Automatic SNN Algorithm on Big Data Sets

Abstract

Clustering is defined as the classification of patterns into groups (clusters) without supervision. The clustering of similarities of data is a complex process that can not be done with human hands. There are various clustering algorithms based on different principles in the literature. The SNN (Shared Nearest Neighborhood) algorithm is a density-based clustering algorithm that identifies similarities between the data by looking at the shared nearest neighbors by two data. The SNN algorithm uses parameters specifying the radius (Eps) that a user enters when clustering, a radius that limits a neighborhood of a point, and the minimum number of points (minPorts) that must be in an eps-neighborhood. This leads to clustering performans has dependency of user experience. A rule-based automatic SNN algorithm has been proposed to remove this dependency from the user. In this study, the performance of the rule-based automatic SNN algorithm over the data sets with 2000 and over sample numbers is examined and presented.