• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2006.05a
• /
• pp.305-312
• /
• 2006

K-means clustering algorithm 에서 주로 이루어지는 랜덤 초기화 (random initialization) 방법은 전역 최적화된 해(global minimum)를 찾아내기에 문제점을 지니고 있다. 즉, 여러 횟수의 알고리듬 반복(iteration)을 실행하더라도 전역 최적화된 해를 찾아내기가 매우 힘들며 주어진 자료의 크기(data size)가 큰 경우에 있어서 이는 거의 불가능하다. 본 논문은 이러한 문제점들을 극복하기 위한 방안으로, wavelet을 이용하여 최적의 초기 군집 중심점(initial clustering center)들을 선택하는 방법을 제시한다. 즉, 웨이블릿을 이용한 효과적인 초기화 (initialization)를 통해서 작은 알고리듬 반복 횟수만으로도 전역 최적화에 도달하는 초기화 방법을 기술한다. 이런 초기화 방법이 군집 알고리즘에 사용될 경우, 온라인상에서 실시간 이루어지는 군집 분석에 큰 도움이 된 수 있다.

• The Journal of the Korea Contents Association
• /
• v.9 no.1
• /
• pp.106-114
• /
• 2009

In this paper we proposed a fast method for a K-Means Clustering algorithm. The main characteristic of this method is that it uses precalculated data which possibility of change is high in order to speed up the algorithm. When calculating distance to cluster centre at each stage to assign nearest prototype in the clustering algorithm, it could reduce overall computation time by selecting only those data with possibility of change in cluster is high. Calculation time is reduced by using the distance information produced by K-Means algorithm when computing expected input data whose cluster may change, and by using such distance information the algorithm could be less affected by the number of dimensions. The proposed method was compared with original K-Means method - Lloyd's and the improved method KMHybrid. We show that our proposed method significantly outperforms in computation speed than Lloyd's and KMHybrid when using large size data which has large amount of data, great many dimensions and large number of clusters.

• Journal of the Semiconductor & Display Technology
• /
• v.23 no.3
• /
• pp.21-28
• /
• 2024

Unsupervised learning is a type of machine learning, and unlike supervised learning or reinforcement learning, a target value for input value is not given. Clustering is mainly used for such unsupervised learning. One of the representative methods of such clustering is K-means clustering. Since K-means clustering is a method of determining the number of clusters and continuing to find the central point of the data allocated to the cluster, there is a problem that the clustered group may not be the optimal cluster. In this study, particle swarm optimization algorithm, which determines the motion vector by adding various variables as well as the center point, is applied to K-means clustering. The improved K-means clustering makes it possible to move toward better outcome values even when the center of cluster no longer change. In the conventional clustering method, the center of the cluster moves to the center of the data belonging to the cluster, and clustering ends when the cluster does not change, so other characteristics other than the center value are excluded. Unlike the conventional clustering method, the improved clustering method uses a central value, an average value, and a random value as variables, and a particle swarm optimization algorithm that modifies the vector for each iteration is applied. As a result, improved clustering method derived a better result value than the existing clustering method in the group's fitness index, silhouette score.

• International Journal of Internet, Broadcasting and Communication
• /
• v.6 no.1
• /
• pp.16-22
• /
• 2014

Clustering methods such as k-means and EM are the group of classification and pattern recognition, which are used in management science and literature search widely. In this paper, k-means and EM algorithm are compared the performance using by Weka. The winning Lottery numbers of 567 cases are experimented for our study and presentation. Processing speed of the k-means algorithm is superior to the EM algorithm, which is about 0.08 seconds faster than the other. As the result it is summerized that EM algorithm is better than K-means algorithm with comparison of accuracy, precision and recall. While K-means is known to be sensitive to the distribution of data, EM algorithm is probability sensitive for clustering.

• Journal of Korea Multimedia Society
• /
• v.21 no.8
• /
• pp.952-959
• /
• 2018

Clustering is one of the most fundamental algorithms in machine learning. The performance of clustering is affected by the distribution of data, and when there are more data or more dimensions, the performance is degraded. For this reason, we use a stacked auto encoder, one of the deep learning algorithms, to reduce the dimension of data which generate a feature vector that best represents the input data. We use k-means, which is a famous algorithm, as a clustering. Sine the feature vector which reduced dimensions are also multi dimensional, we use the Euclidean distance as well as the cosine similarity to increase the performance which calculating the similarity between the center of the cluster and the data as a vector. A deep clustering networks combining a stacked auto encoder and k-means re-trains the networks when the k-means result changes. When re-training the networks, the loss function of the stacked auto encoder and the loss function of the k-means are combined to improve the performance and the stability of the network. Experiments of benchmark image ad document dataset empirically validated the power of the proposed algorithm.

• Journal of Creative Information Culture
• /
• v.4 no.3
• /
• pp.221-230
• /
• 2018

Collaborative filtering technique is a major method of recommender systems and has been successfully implemented and serviced in real commercial online systems. However, this technique has several inherent drawbacks, such as data sparsity, cold-start, and scalability problem. Clustering-based collaborative filtering has been studied in order to handle scalability problem. This study suggests a collaborative filtering system which utilizes genetic algorithms to improve shortcomings of K-means algorithm, one of the widely used clustering techniques. Moreover, different from the previous studies that have targeted for optimized clustering results, the proposed method targets the optimization of performance of the collaborative filtering system using the clustering results, which practically can enhance the system performance.

• Journal of KIISE:Software and Applications
• /
• v.37 no.3
• /
• pp.214-221
• /
• 2010

The search-results clustering problem is defined as the automatic and on-line grouping of similar documents in search results returned from a search engine. In this paper, we propose a new search-results clustering algorithm specialized for a paper search service. Our system consists of two algorithmic phases: Category Hierarchy Generation System (CHGS) and Paper Clustering System (PCS). In CHGS, we first build up the category hierarchy, called the Field Thesaurus, for each research field using an existing research category hierarchy (KOSEF's research category hierarchy) and the keyword expansion of the field thesaurus by a word clustering method using the K-means algorithm. Then, in PCS, the proposed algorithm determines the category of each paper using top-down and bottom-up methods. The proposed system can be used in the application areas for retrieval services in a specialized field such as a paper search service.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.4
• /
• pp.1773-1795
• /
• 2020

Unmanned aerial vehicles (UAVs) network are a very vibrant research area nowadays. They have many military and civil applications. Limited bandwidth, the high mobility and secure communication of micro UAVs represent their three main problems. In this paper, we try to address these problems by means of secure clustering, and a security clustering algorithm based on integrated trust value for UAVs network is proposed. First, an improved the k-means++ algorithm is presented to determine the optimal number of clusters by the network bandwidth parameter, which ensures the optimal use of network bandwidth. Second, we considered variables representing the link expiration time to improve node clustering, and used the integrated trust value to rapidly detect malicious nodes and establish a head list. Node clustering reduce impact of high mobility and head list enhance the security of clustering algorithm. Finally, combined the remaining energy ratio, relative mobility, and the relative degrees of the nodes to select the best cluster head. The results of a simulation showed that the proposed clustering algorithm incurred a smaller computational load and higher network security.

• Knowledge Management Research
• /
• v.10 no.4
• /
• pp.17-30
• /
• 2009

This study is concerned with suggesting a new clustering algorithm to evaluate the value of papers which were supported by research grants by Korea Research Fund (KRF). The algorithm is based on an extended version of a conventional PSO (Particle Swarm Optimization) mechanism. In other words, the proposed algorithm is based on integration of k-means algorithm and simulated annealing mechanism, named KASA-PSO. To evaluate the robustness of KASA-PSO, its clustering results are evaluated by research grants experts working at KRF. Empirical results revealed that the proposed KASA-PSO clustering method shows improved results than conventional clustering method.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.06a
• /
• pp.148-152
• /
• 1998

Gustafson and Kessel proposed a modified fuzzy c-Means algorithm based of the Mahalanobis distance. Though the algorithm appears more natural through the use of a fuzzy covariance matrix, it needs to calculate determinants and inverses of the c-fuzzy scatter matrices. This paper proposes a fuzzy clustering algorithm using pseudo mahalanobis distance, which is more easy to use and flexible than the Gustafson and Kessel's fuzzy c-Means.