Comparing K-means, K-medoids and ISODATA clustering algorithms for a cloud service search engine

Abstract

The evolution of cloud computing over the past few years is potentially one of the major advances in the history of computing. Cloud computing theoretically provides all computing needs as services. Accordingly, a large number of cloud service providers exist and the number is constantly increasing. This presents a significant problem for a user to find a relevant service provider, and calls for developing a specialized search engine to help users select suitable services matching their needs. Towards this goal, we developed a search engine that crawls the web sites of various service providers, extracts service attributes from their JavaScript Object Notation (JSON) files and normalizes the attributes in a service table. Those attributes are clustered using one of three different algorithms (K-means, K-medoids, and ISODATA). The requirements of a given user are then matched against the centroids of the various clusters to help obtain the closest match. In this paper, we compared the three algorithms with respect to time and accuracy. The ISODATA algorithm exhibited the best performance.