Data clustering using the Bees Algorithm and the Kd-Tree structure

Data clustering has been studied intensively during the past decade. The K-means and C-means algorithms are the most popular of clustering techniques. The former algorithm is suitable for 'crisp' clustering and the latter, for 'fuzzy' clustering. Clustering using the K-means or C-means algorithms generally is fast and produces good results. Although these algorithms have been successfully implemented in several areas, they still have a number of limitations. The main aim of this work is to develop flexible data management strategies to address some of those limitations and improve the performance of the algorithms. The first part of the thesis introduces improvements to the K-means algorithm. A flexible data structure was applied to help the algorithm to find stable results and to decrease the number of nearest neighbour queries needed to assign data points to clusters. The method has overcome most of the deficiencies of the K-means algorithm. The second and third parts of the thesis present two new clustering algorithms that are capable of locating near optimal solutions efficiently. The proposed algorithms combine the simplicity of the K-means algorithm and the C-means algorithm with the capability of a new optimisation method called the Bees Algorithm to avoid local optima in crisp and fuzzy clustering, respectively. Experimental results for different data sets have demonstrated that the new clustering algorithms produce better performances than those of other algorithms based upon combining an evolutionary optimisation tool and the K-means and C-means clustering methods. The fourth part of this thesis presents an improvement to the basic Bees Algorithm by applying the concept of recursion to reduce the randomness of its local search procedure. The improved Bees Algorithm was applied to crisp and fuzzy data clustering of several data sets. The results obtained confirm the superior performance of the new algorithm