K-Means Clustering using Nature-Inspired Optimization Algorithms-A Comparative Survey

Abstract

In the past few decades, an in depth and extensive research has been administered on K-Means to
combine with genetic and nature-inspired optimization algorithms for clustering. As conventional KMeans have drawbacks of stalling out at nearby optima which is needy upon the arbitrary estimations
of the underlying centers of the clusters. Optimization algorithms usually search the entire search space
for an optimal solution by maintaining a strategic distance from the neighborhood optima. These help
to accelerate the clustering procedure with various searching algorithms incorporating Firefly, wolf,
ANT, Cuckoo and BAT are some of the strategies that the optimization algorithms use to move quicker
to find the ideal solution. These purported bio-enlivened techniques have advantages and disadvantages
based on the attributes they use in their search behavior. In this paper a detailed study of K-Means
combined with optimization algorithm along with their performance metrics such as network lifetime,
energy efficiency, intra-cluster distances, F-measure, accuracy and so on are explored. Findings
specify that Whale Optimization algorithm alone gives better results in almost all the metrics specified
in wireless Sensor Networks but did not excel in parameters of clustering such as intra-cluster
distances, F-measure and accuracy. So, after keen study a WOA combined with K-Means clustering to
select cluster heads in a Wireless Sensor Network is initiated for improvements in all the metrics.
Index Terms- Selection of cluster head, K-Means, Firefly, ACO Algorithm, PSO algorithm, Whale
Optimization Algorithm, Wireless Sensor Networks (WSN), Network Lifetime.