Mining of projected clusters in high dimensional data using modified fuzzy C means algorithm

Abstract

Clustering is an important task in data mining. Data mining is the newlineprocess of extracting useful and hidden information from huge amount of newlinedata, which is generated everywhere. It is highly difficult to identify and newlinelocate useful information from the huge data. Data mining algorithms face the newlinechallenge of handling huge data, different types of data such as numeric, newlinecategorical, spatial, multimedia data and so on, reporting most interesting data newlineof reasonable sizes so that users can interpret them and also to make use of newlinevisualization techniques to provide the results to users in a more newlineunderstandable way. Various data mining techniques are available such as newlineclassification, association rule mining, pattern recognition and clustering etc. newlineClustering is of research interest in this thesis.Clustering is a process of grouping similar objects together. In newlinegeneral, grouping of objects is required in many applications such as customer segmentation, trend analysis and classification. Most of the newlineclustering algorithms do not perform well in high dimensional data because of newlineirrelevant dimensions. In high dimensional space, the distance between a pair newlineof points is less precise as the number of dimensions increases. Clusters can newlinewell be formed in some of the projected dimensions of the data space. newlineClustering algorithms can be classified as partitional, hierarchical, newlinedensity based, model based, grid based and projected clustering algorithms. In newlinethis approach, a new projected clustering algorithm is invented which is newlinepartitional in nature. newlineProjected clustering is a branch of clustering which is receiving newlinemore attention from database communities nowadays. Projected clustering newlinecan be defined as finding clusters and their relevant dimensions. In normal newlineclustering algorithms clusters are formed by grouping objects based on their newlinedistance in all dimensions. In the resultant clusters all dimensions are newlineincluded. But in the projected clustering, only relevant dimensions are newlineincluded in the resultant clusters. newline newline