An evolutionary computing framework to cluster heterogeneous data by incorporating fusing of attributes based on importance

Abstract

newlineClustering is a type of unsupervised learning method and it is used to group the data objects The similar objects are put into same group and the dissimilar objects in different groups The foremost property of any clustering technique is to maximize the intra cluster similarity and minimize the inter cluster similarity In today s world of digital era and due to the development of technology all the real world entities are generating data and those data are often heterogeneous in nature There are many techniques available for clustering homogeneous data and those techniques do not perfectly suit to cluster heterogeneous data The size dimensionality attribute domains number of attributes order and structure are the major problems associated with clustering heterogeneous data In this thesis the main focus is to consider the wider imensionality and number of attributes The first contribution adopts the advantages of Genetic Algorithm GA for identifying the important attribute subsets In the second contribution individual distance measures are defined for heterogeneous data over different data types namely numeric binary nominal and ordinal and then interblend fusing of distance matrix is proposed The third work proposes the use of single interblend fusing of distance matrix with various partition clustering algorithm In the fourth work the partition clustering newline