Article Preview
Top1. Introduction
Data mining is the process of discovering useful patterns and knowledge from large volume of data (Han et al., 2011). Primarily, four types of learning approaches have been described in data mining such as supervised, unsupervised, semi-supervised and active user learning. The supervised learning analyzes the data object with respect to class labels, while, unsupervised learning analyzes the data object without consulting the class labels. The semi-supervised learning considers both supervised and unsupervised learning approaches. Whereas, in active-user learning, user can interactively label the data points with desired outputs. Clustering is an unsupervised machine learning approach that divides a set of objects into distinct clusters (Jain et al., 1999; Nanda & Panda 2014). The objects within a cluster are similar to each other and dissimilar to other clusters. The prime objective of clustering is to maximize the intra-cluster compactness and minimizing inter-cluster likeness among clusters. Last few decades, momentous research has been done in the clustering field. Several optimization techniques inspired from natural phenomenon have been reported to obtain optimal solutions for clustering task. Some of these are particle swarm optimization (Cura 2012), Magnetic optimization algorithm (Kushwaha et al., 2018), charged system search approach (Kumar & Sahoo 2014), Black hole (Hatamlou 2013), artificial bee colony algorithm (Karaboga & Ozturk 2011), ant colony optimization (Shelokar et al., 2004) and big bang big crunch algorithm (Hatamlou et al., 2011).
In recent time, a heuristic algorithm based on cat intelligence has gained wide popularity among research community and adopted in several research fields like workflow scheduling in the cloud, image analysis, wireless sensor networks, data analysis etc. (Chu et al., 2006; Tsai et al., 2012; Ram et al., 2015; Wang et al., 2012; ). Initially, Santosa and Ningrum (2009) applied the CSO algorithm for solving clustering problems. This algorithm works in two modes, seeking mode and tracing mode. The resting behavior of cats is described using seeking mode, while the hunting skill of cat is described using tracing mode. The seeking mode responsible for local search, whereas the tracing mode responsible for global search. It is observed that CSO algorithm have good exploration capability, but suffers from weak exploitation ability (Kumar & Sahoo 2017). Sometimes, the CSO algorithm cannot explore entire search space for an optimum solution due to lack of global best position information and results in slow convergence rate (Kumar & Singh 2018). These deficiencies can affect the performance of CSO algorithm for solving optimization problems. Hence, to make the CSO algorithm more efficient and robust, several issues are identified and resolved in this research work.