Mapreduce based partitional clustering algorithms for handling large scale data

Abstract

Data mining is the process of finding the hidden patterns in data according to different perspectives and inventing useful information from these patterns. It constitutes some techniques that facilitate the decision making and other information requirements to take the right decisions at the right time. Cluster analysis plays a major role in data mining techniques and discover the meaningful patterns without prior knowledge about the data. The topmost categories of clustering algorithms are partitional and hierarchical. The partitional clustering algorithms form the clusters by dividing the data objects into groups while hierarchical clustering algorithms form the clusters y the hierarchical decomposition of data objects. The K-Means is the most popular and widely used partitional clustering algorithm due to its simplicity and performance. However, this algorithm is primarily useful for clustering the numerical data only. It is extended to group the categorical and mixed numeric and categorical types of data. These algorithms are called K-Modes and K-Prototypes clustering algorithms. In this study, the term partitional clustering algorithms specify the K-Means, K-Modes and K-Prototypes algorithms. These clustering algorithms select the initial centroids randomly. Due to this nature, these algorithms provides the clustering solutions with local optima and very poor quality of clusters. Similarly, the main objective of these clustering algorithms is to minimize the distance between data instances and their cluster centroids. To handle problems with these clustering algorithms by utilizing the natureinspired optimization algorithms. newline