Towards Developing Fuzzy Neighborhood Based Clustering Algorithms for High Performance Distributed Memory Computing Environments

Abstract

Fuzzy neighborhood-based clustering algorithms overcome the parameter selection problem of classical neighborhood based clustering algorithms and offer fully unsupervised, i.e., parameter free clustering. On the other hand, due to the inherent fuzzy-calculation-overhead, they demand higher processing time and memory compared to classical clustering algorithms. In some recent studies, these fuzzy algorithms have been improved, especially in terms of speed, such that they became applicable to large data sets. Nonetheless, they need to be adapted to multi-computer systems in order to be used in today's big data applications. The aim of this study is developing fuzzy neighborhood-based clustering algorithms which are designed to run on high performance distributed memory computing environments and revealing their effectiveness by testing them in a real big-data application.