Recent Development of Metaheuristics for Clustering

Abstract

Abstract. Metaheuristics have been successfully applied to quite a lot of services, systems, and products frequently found in our daily life. Until now, none of the metaheuristics ever proposed are perfect for all the optimization problems; rather, each algorithm has its pros and cons. Although several highperformance metaheuristics exist, there is still plenty of room to improve the final result they produce and the computation time they take. Since 2001, quite a few number of novel metaheuristics have been developed to provide a better way for solving the optimization problems. A brief review for eight of these novel metaheuristics is given in this paper. To evaluate the performance of these algorithms, we apply them to a well-known combinatorial optimization problem, data clustering, and the results are analyzed and discussed.

Big data analytics: a survey

“…For example, several studies [114,145] used k-means as an example to analyze the big data, but not many studies applied the state-of-the-art data mining algorithms and machine learning algorithms to the analysis the big data. This explains that the performance of the big data analytics can be improved by data mining algorithms and metaheuristic algorithms presented in recent years [147]. The relevant technologies for compression, sampling, or even the platform presented in recent years may also be used to enhance the performance of the big data analytics system.…”

A Multi-Objective Gwo Based Clustering Approach for Efficient Document Mining

“…As digitization techniques improve, a considerable portion of all written data is now kept digitally (as soft copies). Document clustering is thus one of the most essential uses, and it is becoming increasingly relevant [4]. A fast and efficient clustering method was required to get valuable data from a big database.…”

“…Traditional clustering approaches, such as k-means clustering, rely heavily on the initial choice of cluster center, which must be rerun many times to yield the best results. These problems can be solved by treating the clustering as an optimization problem, much like, the clustering problem is generally defined as: Given a set of n patterns X = {x 1 , x 2 , … x n } in d dimensional space, partition the set X into k clusters C = {c 1 , c 2 , … c k }that minimize a predetermined criterion (for example, sum of squared errors (SSE), entropy, f-measure, or accuracy) [4]. The advantage of meta heuristic clustering over classical clustering is that the former is unaffected by starting cluster locations and may be easily adjusted by user-defined objective functions.…”

Clustering based on the Krill Herd Algorithm with Selected Validity Measures

“…The food coefficient in (16), expresses the global attraction of the food centre (15), and may be calculated as:…”

“…Yet one more group of clustering algorithms is that based on an optimization algorithm inspired by Nature [16], [17]. In this approach, some metaheuristics are applied for the optimization of adopted division criteria.…”