No abstract

Abstract— When data sets have one or more similar characteristics, the clustering in each of these data sets will have an effect on the other data sets. However, for various reasons such as data security issues, these data cannot be stored centrally but in different places. Collaborative clustering is a clustering technique that allows to performance of local clustering on each sub-data set and to exchange of information with other data sets. A collaborative process will be performed to adjust the clustering results on each subset to achieve better clustering results on the subsets of data. This paper presents a collaborative fuzzy clustering approach in big data analysis based on a high-performance computational model to improve the computation speed. Experiments on the Kitsune network attack dataset show that the proposed algorithm significantly improves the calculation speed compared to the previous method.

Possibilistic Fuzzy c-means (PFCM) algorithm is a powerful clustering algorithm. It is a combination of two algorithms Fuzzy c-means (FCM) and Possibilistic c-means (PCM). PFCM algorithm deals with the weaknesses of FCM in handling noise sensitivity and the weaknesses of PCM in the case of coincidence clusters. However, PFCM still has a common weakness of clustering algorithms that is easy to fall into local optimization. Cuckoo search (CS) is a novel evolutionary algorithm, which has been tested on some optimization problems and proved to be stable and high-efficiency. In this study, we propose a hybrid method encompassing PFCM and improved Cuckoo search to form the proposed PFCM-ICS. The proposed method has been evaluated on 4 data sets issued from the UCI Machine Learning Repository and compared with recent clustering algorithms such as FCM, PFCM, PFCM based on particle swarm optimization (PSO), PFCM based on CS. Experimental results show that the proposed method gives better clustering quality and higher accuracy than other algorithms.