• Journal of the Korean Data and Information Science Society
• /
• v.21 no.6
• /
• pp.1353-1359
• /
• 2010

Basic objective in cluster analysis is to discover natural groupings of items. In general, clustering is conducted based on some similarity (or dissimilarity) matrix or the original input data. Various measures of similarities between objects are developed. In this paper, we consider a clustering of huge categorical real data set which shows the aspects of time-location-activity of Korean people. Some useful similarity measure for the data set, are developed and adopted for the categorical variables. Hierarchical and nonhierarchical clustering method are applied for the considered data set which is huge and consists of many categorical variables.

• The Korean Journal of Applied Statistics
• /
• v.26 no.5
• /
• pp.821-834
• /
• 2013

Law et al. (2004) proposed a normal distribution based salient mixture model for variable selection in clustering. However, this model has substantial problems such as the unidentifiability of components an the inaccurate selection of informative variables in the case of a small cluster size. We propose an alternative method to overcome problems and demonstrate a good performance through experiments on simulated data and real data.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2000.04a
• /
• pp.601-604
• /
• 2000

In this paper, we compare the classification performance of both data fusion and clustering algorithms (Data Bagging, Variable Selection Bagging, Parameter Combining, Clustering) to logistic regression in consideration of various characteristics of input data. Four factors used to simulate the logistic model are (1) correlation among input variables (2) variance of observation (3) training data size and (4) input-output function. Since the relationship between input & output is not typically known, we use Taguchi design to improve the practicality of our study results by letting it as a noise factor. Experimental study results indicate the following: Clustering based logistic regression turns out to provide the highest classification accuracy when input variables are weakly correlated and the variance of data is high. When there is high correlation among input variables, variable bagging performs better than logistic regression. When there is strong correlation among input variables and high variance between observations, bagging appears to be marginally better than logistic regression but was not significant.

• The Bulletin of The Korean Astronomical Society
• /
• v.36 no.2
• /
• pp.131.1-131.1
• /
• 2011

We present applications of clustering methods to detect variability in massive astronomical time series data. Focusing on variability of bright stars, we use clustering methods to separate possible variable sources from other time series data, which include intrinsically non-variable sources and data with common systematic patterns. We already finished the analysis of the Northern Sky Variability Survey data, which include about 16 million light curves, and present candidate variable sources with their association to other data at different wavelengths. We also apply our clustering method to the light curves of bright objects in the SuperWASP Data Release 1. For the analysis of the SuperWASP data, we exploit a elastically configurable Cloud computing environments that the KISTI Supercomputing Center is deploying. Two quite different configurations are incorporated in our Cloud computing test bed. One system uses the Hadoop distributed processing with its distributed file system, using distributed processing with data locality condition. Another one adopts the Condor and the Lustre network file system. We present test results, considering performance of processing a large number of light curves, and finding clusters of variable and non-variable objects.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.4
• /
• pp.485-492
• /
• 2009

In the mobile service environment, patterns generated by temporal synchronization are streamed with different instance values. This paper proposed a variable clustering management method, which manages multiple data streaming dynamically, to support flexible clustering. The method manages synchronization effectively and differently with conventional streaming methods in data streaming environment and manages clustering streaming after the structural presentation level and the fitness presentation level. In the structural presentation level, the stream structure is presented using level matching and accumulation matching, and clustering management is carried out by the management of dynamic segment and static segment. The performance of the proposed method is tested by using k-means method, C/S server method, CDN method, and simulation. The test results showed that the proposed method has better performance than the other methods.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.10b
• /
• pp.233-237
• /
• 2006

This study proposes an innovative measure for evaluating the performance of text clustering. In using K-means algorithm and Kohonen Networks for text clustering, the number clusters is fixed initially by configuring it as their parameter, while in using single pass algorithm for text clustering, the number of clusters is not predictable. Using labeled documents, the result of text clustering using K-means algorithm or Kohonen Network is able to be evaluated by setting the number of clusters as the number of the given target categories, mapping each cluster to a target category, and using the evaluation measures of text. But in using single pass algorithm, if the number of clusters is different from the number of target categories, such measures are useless for evaluating the result of text clustering. This study proposes an evaluation measure of text clustering based on intra-cluster similarity and inter-cluster similarity, what is called CI (Clustering Index) in this article.

• International Journal of Control, Automation, and Systems
• /
• v.4 no.4
• /
• pp.438-447
• /
• 2006

Fuzzy controller's design depends mainly on the rule base and membership functions over the controller's input and output ranges. This paper presents two different approaches to deal with these design issues. A simple and efficient approach; namely, Fuzzy Subtractive Clustering is used to identify the rule base needed to realize Fuzzy PI and PD type controllers. This technique provides a mechanism to obtain the reduced rule set covering the whole input/output space as well as membership functions for each input variable. But it is found that some membership functions projected from different clusters have high degree of similarity. The number of membership functions of each input variable is then reduced using a similarity measure. In this paper, the fuzzy subtractive clustering approach is shown to reduce 49 rules to 8 rules and number of membership functions to 4 and 6 for input variables (error and change in error) maintaining almost the same level of performance. Simulation on a wide range of linear and nonlinear processes is carried out and results are compared with fuzzy PI and PD type controllers without clustering in terms of several performance measures such as peak overshoot, settling time, rise time, integral absolute error (IAE) and integral-of-time multiplied absolute error (ITAE) and in each case the proposed schemes shows an identical performance.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.13 no.12
• /
• pp.2603-2608
• /
• 2009

Document clustering is an important method for document analysis and is used in many different information retrieval applications. This paper proposes a new document clustering model using the clustering method based NMF(non-negative matrix factorization) and refinement of documents in cluster by using coherence of cluster. The proposed method can improve the quality of document clustering because the re-assigned documents in cluster by using coherence of cluster based similarity between documents, the semantic feature matrix and the semantic variable matrix, which is used in document clustering, can represent an inherent structure of document set more well. The experimental results demonstrate appling the proposed method to document clustering methods achieves better performance than documents clustering methods.

• Proceedings of the KSRS Conference
• /
• 2008.10a
• /
• pp.111-114
• /
• 2008

The existence of noise bands may deform the typical shape of the spectrum, making the accuracy of clustering degraded. This paper proposes a statistical approach to remove noise bands in hyperspectral data using the correlation coefficient of bands as an indicator. Considering each band as a random variable, two adjacent signal bands in hyperspectral data are highly correlative. On the contrary, existence of a noise band will produce a low correlation. For clustering, the unsupervised ${\kappa}$-nearest neighbor clustering method is implemented in accordance with three well-accepted spectral matching measures, namely ED, SAM and SID. Furthermore, this paper proposes a hierarchical scheme of combining those measures. Finally, a separability assessment based on the between-class and the within-class scatter matrices is followed to evaluate the applicability of the proposed noise band removal method. Also, the paper brings out a comparison for spectral matching measures.

• The Korean Journal of Applied Statistics
• /
• v.13 no.2
• /
• pp.343-352
• /
• 2000

In this study. the author proposes a nonhierarchical clustering method. called the "Double K-Means Clustering", which performs clustering of multivariate observations with the following algorithm: Step I: Carry out the ordinary K-means clmitering and obtain k temporary clusters with sizes $n_1$,... , $n_k$, centroids $c_$1,..., $c_k$ and pooled covariance matrix S. $\bullet$ Step II-I: Allocate the observation x, to the cluster F if it satisfies ..... where N is the total number of observations, for -i = 1, . ,N. $\bullet$ Step II-2: Update cluster sizes $n_1$,... , $n_k$, centroids $c_$1,..., $c_k$ and pooled covariance matrix S. $\bullet$ Step II-3: Repeat Steps II-I and II-2 until the change becomes negligible. The double K-means clustering is nearly "optimal" under the mixture of k multivariate normal distributions with the common covariance matrix. Also, it is nearly affine invariant, with the data-analytic implication that variable standardizations are not that required. The method is numerically demonstrated on Fisher's iris data.