An unsupervised grid-based approach for clustering analysis

Abstract

In recent years, the growing volume of data in numerous clustering tasks has greatly boosted the existing clustering algorithms in dealing with very large datasets. The K-means has been one of the most popular clustering algorithms because of its simplicity and easiness in application, but its efficiency and effectiveness for large datasets are often unacceptable. In contrast to the K-means algorithm, most existing grid-clustering algorithms have linear time and space complexities and thus can perform well for large datasets. In this paper, we propose a grid-based partitional algorithm to overcome the drawbacks of the K-means clustering algorithm. This new algorithm is based on two major concepts: 1) maximizing the average density of a group of grids instead of minimizing the minimal square error which is applied in the K-means algorithm, and 2) using gridclustering algorithms to thoroughly reformulate the object-driven assigning in the K-means algorithm into a new grid-driven assigning. Consequently, our proposed algorithm obtains an average speed-up about 10-100 times faster and produces better partitions than those by the K-means algorithm. Also, compared with the K-means algorithm, our proposed algorithm has ability to partition any dataset when the number of clusters is unknown. The effectiveness of our proposed algorithm has been demonstrated through successfully clustering datasets with different features in comparison with the other three typical clustering algorithms besides the K-means algorithm.

Visual analytics for the clustering capability of data

“…To clean the boundary, we first calculate the energy of each data point according to (2), and remove the M 1 data points with minimal AE. We then re-calculate the AE for each data point and again delete http://engine.scichina.com/doi/10.1007/s11432-013-4832-7 M data point, repeating these steps until significant boundaries are evident.…”

“…Cluster analysis is a process for structuring and reducing data sets by finding groups of similar data elements [1,2]. With the development of database system, large scale, and multidimensional datasets are widely available in various fields.…”

Clustering mechanism for electric tomography imaging

“…The correct number of clusters is three according to Eq. (25). Figure 3 (b) and (e) shows the images obtained by the Landweber algorithm while (c) and (f) obtained by the LBP algorithms, respectively.…”

“…The quality of the ET images can be quantitatively assessed. In addition, fuzzy clustering has been widely adopted in industry [25][26][27].…”

Application of grid-based C-means clustering algorithm for image segmentation

“…In order to overcome the above problems, in the recent twenty years some research reports have published [5] [6]. First in this paper we propose a grid-base C-means (G-C-MEANS) algorithm for clustering [7]. The G-C-MEANS algorithm can overcome those problems such as the determination of the number of cluster, determining precise positions of the clustering prototypes, the real-time performances and so on.…”