Design New Meta Heuristic Algorithms for Partitional Clustering Problems

Abstract

In this competitive business world, data mining is an essential aspect of the knowledge discovery process. Clustering is an important data analysis and well-recognized technique in the field of data mining. In clustering, there is no need for training the data. This technique adopts a distance measure to compute data objects in clusters. The implication of clustering techniques in different disciplines primes momentous research in this field. In the present time, clustering can get wide attention from the research community both of theoretical and practical point. It is seen that much work is reported on clustering methods in literature; still, the clustering is an active area of research. This research work focuses on partitional clustering and its disciplines. The partitional clustering divides the data objects into different groups called clusters. Data objects within a cluster are similar in nature and exhibit heterogeneity with other clusters. There are several shortcomings associated with clustering methods like identification of initial cluster centers, stuck in local optima, lack of population diversification mechanism, imbalance exploration and exploitation processes, convergence rate and accuracy. This work addresses the local optima, diversity, convergence rate and exploration and exploitation issues. In this thesis, three algorithms are designed to address the issues as mentioned. An artificial chemical reaction optimization is explored to solve clustering problems. The artificial chemical reaction algorithm has proved its competency in clustering filed and provided good candidate solutions. To address the diversity and convergence rate issue, an improved version of the big bang big crunch algorithm is also proposed. The chaotic maps and cellular automata-based heat transfer methods are introduced in the big bang and big crunch algorithm. Further, to address the local optima and exploration and issue a neighborhood-based cat swarm optimization algorithm is also developed. The position and velocity se