Density-based Algorithms for Big Data Clustering Using MapReduce Framework: A Comprehensive Study

摘要

Clustering is used to extract hidden patterns and similar groups from data. Therefore, clustering as a method of unsupervised learning is a crucial technique for big data analysis owing to the massive number of unlabeled objects involved. Density-based algorithms have attracted research interest, because they help to better understand complex patterns in spatial datasets that contain information about data related to co-located objects. Big data clustering is a challenging task, because the volume of data increases exponentially. However, clustering using MapReduce can help answer this challenge. In this context, density-based algorithms in MapReduce have been largely investigated in the past decade to eliminate the problem of big data clustering. Despite the diversity of the algorithms proposed, the field lacks a structured review of the available algorithms and techniques for desirable partitioning, local clustering, and merging. This study formalies the problem of density-based clustering using MapReduce, proposes a taxonomy to categorize the proposed algorithms, and provides a systematic and comprehensive comparison of these algorithms according to the partitioning technique, type of local clustering, merging technique, and exactness of their implementations. Finally, the study highlights outstanding challenges and opportunities to contribute to the field of density-based clustering using MapReduce.