An empirical comparative analysis of clustering algorithms for big data applications

• Main Author: Delos Santos, Duke Danielle T.
• Format: text
• Language: English
• Published: Animo Repository 2017
• Subjects: Big data; Algorithms
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/5395
• Institution: De La Salle University

Big data is a vaguely defined term that describes a dataset as either too large or too complex to analyze and get satisfactory results. Clustering algorithms are a possible solution to this problem of big data, where they can be categorized according to one or more of three clustering objectives. These are defined as either grouping focused algorithms, in which the algorithm aims to classify the dataset into meaningful groups, data summarization algorithms, in which the algorithm aims to summarize the data point into a more concise format for an easier analysis, and finally, data visualization, in which the dataset is visualized in a more understandable format. While there are only three categories one can classify clustering algorithms, there are a large number of clustering algorithms with differing performances for different sizes of datasets. The algorithms empirically evaluated and compared under the research include k-means, SOM, DBSCAN, BFR, and BIRCH, and it was found that the algorithms all have different strengths and weaknesses when classifying scaled up datasets, and one can choose the appropriate algorithm based on these strengths and weaknesses.