The State Polytechnics of Malang,

Cluster analysis aims at identifying groups of similar objects and, therefore helps to discover distribution of patterns and interesting correlations in the data sets. In this paper, we propose to provide a consistent partitioning of a dataset which allows identifying any shape of cluster patterns in case of numerical clustering, convex or non-convex. The method is based on layered structure representation that be obtained from measurement distance and angle of numerical data to the centroid data and based on the iterative clustering construction utilizing a nearest neighbor distance between clusters to merge. Encourage result show the effectiveness of the proposed technique.


I. INTRODUCTION
Clustering is one of the most useful tasks in data mining process for discovering groups and identifying interesting distributions and patterns in the underlying data and is also broadly recognized as a useful tool in many applications.It has been subject of wide research since it arises in many application domains in engineering, business and social sciences.Researchers of many disciplines have addressed the clustering problem.The objective of algorithms is to minimize the distance of the objects within a cluster from the representative point of this cluster [1].The clustering means process to define a mapping, from some data D={t1, t2,t3,…tn} to some clusters C={c1, c2,c3,…, cn} based on similarity between ti.
Clustering is a central task for which many algorithms have been proposed.The task of finding a good cluster is very critical issues in clustering [2].Cluster analysis constructs good clusters when the members of a cluster have minimize distances (Intra-cluster distances are minimized or internal homogeneity) are also not like members of other clusters (Inter-cluster distances are maximized).Clustering algorithms can be hierarchical or partitional [3].
In hierarchical clustering, the output is a tree showing a sequence of clustering with each cluster being a partition of the data set.Hierarchical clustering proceeds successively by either merging smaller clusters into larger ones.The result of the algorithm is a tree of cluster, called dendogram, which shows how the cluster are related, by cutting the dendogram at a desired level, a clustering of data the data items into disjoint groups is obtained [1].The most well known methods for clustering is K-means developed by Mac Queen in 1967.Kmeans is a partition clustering method that separates data into k mutually excessive groups.Partitional clustering attempts to directly decompose the data set into a set of disjoint clusters.By iterative such partitioning, K-means minimizes the sum of distance from each data to its clusters.This method ability to cluster huge data, and also outliers, quickly and efficiently [4].However, K-means algorithm is very sensitive in initial starting points.Because of initial starting points generated randomly, K-means does not guarantee the unique clustering results.However, many of above clustering methods require some additional user specified parameters, such as shape of cluster, optimal number, similirity threshold etc.In this paper, a new simple algorithm of numerical clustering is proposed.The proposed method, in particular, for a shape independent clustering that can also be applied in the case of condensed clustering.

II. PROPOSED CLUSTERING ALGORITHM
In this paper, a new simple algorithm of numerical clustering is proposed.The proposed method, in particular, is for a shape independent clustering.The proposed method can also be applied in the case of condensed clustering.we implemented multi-layer centroid contour distance (mlccd) [5] with some modification to the clustering.
The algorithm is described as follow:

1) Begin with an assumption that every point "n" is it's own cluster c i, where i =1,2,…n 2) Calculate the centroid location 3) Calculate the angle and distance for every point to the centroid 4) Make multi layer centroid contour distance based on step 2 and 3 5) Set i = 1 as initial counter 6) Increment i = i + 1 7) Measure distance between cluster in the location i with cluster in the location i-1 and cluster in the location i+1. 8) Merge two cluster become one cluster base on the a nearest neighbor distance between clusters (see equation 2.4)
as shown in the step 7 www.ijarai.thesai.org9) Repeat from step 6 to step 8 while i< 360 10) Repeat step 5 to step 9 until the required criteria is met Firstly in the step 1, let every point "n" is it's own cluster, if there are n data it mean there are n cluster.Secondly, in the step 2 obtain the centroid by using equation 2.1 in this case every point have contribution to find the centroid location.
Step3 Calculate the distance and angle between centroid to the every point, the angle is obtained by using arctangent of dy/dx, dy is differences y position between position y of every point and position y in the centroid and dx is differences x position between position x of every point and position x in the centroid see equation 2.2.The distance between centroid to the every point of data can be obtained by using equation (2.3).The next step By using the angle of every point and its distance then make the multi layer centroid contour distance (mlccd).The next process set i as counter start from 1 to 360 then calculate distance by using Euclidian distance between every cluster that pointed by i and every cluster that pointed by i-1 and i+1.After the distance was calculated then merge two cluster become one cluster base on the nearest distance.position of the centroid is: Where: angle every point to the centroid is: (2) Where: ( , ) = Position centroid ( ,y) = Position point or data The distance from the centroid to every point is : (3) Where: ( , ) = Position centroid ( ,y) = Position point or data The a nearest neighbor distance between clusters is calculated by using Euclidian distance these equation is commonly used to calculate the distance in case of numerical data sets [6].For two-dimensional dataset, it performs as:

III. EXPERIMENT RESULT
In order to analyze the accuracy of our proposed method we represent error percentage as performance measure in the experiment.It is calculated from number of misclassified patterns and the total number of patterns in the data sets [7] (see the equation 3.1).We compare our method with the conventional clustering methods (k-means and other) to the same dataset. (5) We applied the proposed method to solve some various shape independent cases and also tried to apply the proposed method in condensed clustering case.The dataset consist Circular nested dataset contain 96 data and 3 cluster with cluster1 contain 8 data, cluster 2 contain 32 data and cluster 3 contain 56 data, inter related dataset contain 42 data and 2 cluster with cluster 1 contain 21 data and cluster 2 contain 21 data, S shape dataset contain 54 data and 3 cluster with cluster1 contain 6 data, cluster 2 contain 6 data and cluster 3 contain 42 data.u shape dataset contain 38 data and 2 cluster with cluster 1 contain 12 data and cluster 2 contain 26 data.The 2 cluster Random dataset contain 34 data and 2 cluster with cluster1 contain 15 data and cluster 2 contain 19.The 3 cluster condense dataset contain 47 data and 3 cluster with cluster1 contain 16 data, cluster 2 contain 14 data and cluster 3 contain 17.The last data is 4 cluster condense dataset contain 64 data and 4 cluster with cluster1 contain 14 data, cluster 2 contain 18 data cluster 3 contain 15 data and cluster 4 contain 17 data.In the Figure 1 up to Figure 7, A is unlabelled data, B is labelled data by using proposed method and C is labelled data by using K-mean method.In Figure 4 by using proposed method there is no misclassified, by using k-mean there are misclassified in the cluster1 3 miss and cluster2 11 miss, average error is 33.65%.In Figure 5 by using proposed method and k-mean there is no misclassified average error is 0%.www.ijarai.thesai.orgA B C Fig. 6.The 3 cluster condense dataset contain 47 data and 3 cluster with cluster1 contain 16 data, cluster 2 contain 14 data and cluster 3 contain 17 In Figure 6 and Figure 7 by using proposed method and kmean there is no misclassified average error is 0%.We also implement some other clustering method that are Hierarchical algorithms (Single Linkage, Centroid Linkage, Complete Linkage and Average Linkage) to the same dataset the average result shown in the Table2.In the Table 1 and Table2 are shown that the proposed method compare with other method allows to identifying any shape of cluster as well as condensed dataset.

IV. CONCLUSION
In this work a new clustering methodology has been introduced based on the layered structure representation that be obtained from measurement distance and angle of numerical data to the centroid data and based on the iterative clustering construction utilizing a nearest neighbor distance between clusters to merge.From the experimental results with some various clustering cases, the proposed method can solve the clustering problem and create well-separated clusters.It is found that the proposed provide a consistent partitioning of a dataset which allows identifying any shape of cluster patterns in case of numerical clustering as well as condensed clustering.
position of the centroid in the x axis Y c = position of the centroid in the y axis n = Total point or data (every point have x position and y position)

TABLE I .
AVERAGE ERROR IN PERCENT OF CLUSTERING BY USING LAYERED STRUCTURE REPRESENTATION AND CLUSTERING BY USING K-MEAN