Related work

Traditional Clustering algorithms can be categorized into four groups: partition clustering, hierarchical clustering, density-based clustering and grid-based clustering. In partitioning clustering category, data is split into k partitions (clusters) using an iterative relocation process in order to enhance the similarity of each partition. Hierarchical clustering is the next category where data is clustered in a hierarchical fashion using bottom-up approach or top-down approach. Density-based clustering is another category where clusters are identified using a growing scheme based on a density threshold that neighborhood objects must exceed. Although arbitrary shapes (Ghanem 2015) and (Ghanem 2015) noise can be detected using these clustering methods, they are not scale well with the size of dataset. Finally, Grid-based clustering algorithms partition data space into grid of cells (Ghanem 2015) which are combined to form clusters based on neighborhood relations. It is distinguished by fastness but it does not work efficiently in high dimensional space. Unlike Traditional clustering algorithms, subspace clustering has been proposed to overcome problems arisen from curse of dimensionality phenomena by constructing clusters based on similarities on a subset of attributes (subspace). As a result, some samples may be assigned to multiple clusters. If each sample is assigned to only one cluster based on some subset of attributes, subspace clustering will be called projected clustering.

                                     

12211212

	212	211	212
1212	212		1
12	121	2112	212