A Clustering-Based Algorithm for Data Reduction

Proceedings of the Fifth Balkan Conference in Informatics on - BCI '12, 2012

Although the k-Nearest Neighbor classifier is one of the most widely-used classification methods, it suffers from the high computational cost and storage requirements it involves. These major drawbacks have constituted an active research field over the last decades. This paper proposes an effective data reduction algorithm that has low preprocessing cost and reduces storage requirements while maintaining classification accuracy at an acceptable high level. The proposed algorithm is based on a fast pre-processing clustering procedure that creates homogeneous clusters. The centroids of these clusters constitute the reduced training-set. Experimental results, based on real-life datasets, illustrate that the proposed algorithm is faster and achieves higher reduction rates than three known existing methods, while it does not significantly reduce the classification accuracy.

The inductive learning of fuzzy rule-based classification systems usually encounters an issue of exponential growth of the fuzzy rulesearch space when the number of patternsand/or variables becomes large.This issue makes the learning process more difficultand, in most cases, may lead to scalability problems. Alcalá-Fdezet al.proposed a fuzzy association rule-based classification method for high-dimensional problems, whichis based on three stages to obtain an accurate and compact fuzzyrule-based classifier with a low computational cost.But there is a serious drawback with this method: the initial linguistic termsmust be predefined by the user. Weapply a self-constructing clustering technique for determining the linguistic terms automatically according to thecharacteristics of the training data. Therefore, the resulting classification system can be more friendly and time-saving for use to the user. Furthermore, more accurate classification results can usually be obtained for the user.

2007

We consider the problem of reducing a potentially very large dataset to a subset of representative prototypes. Rather than searching over the entire space of prototypes, we first roughly divide the data into balanced clusters using bisecting k-means and spectral cuts, and then find the prototypes for each cluster by affinity propagation. We apply our algorithm to text data, where we perform an order of magnitude faster than simply looking for prototypes on the entire dataset. Furthermore, our "divide and conquer" approach actually performs more accurately on datasets which are well bisected, as the greedy decisions of affinity propagation are confined to classes of already similar items.

IEEE Transactions on Systems, Man and Cybernetics, Part B (Cybernetics), 1999

2012 26th International Conference on Advanced Information Networking and Applications Workshops, 2012

The k-Nearest-Neighbours(kNN) is a simple and effective method for data classification. One of the major drawbacks of kNN is its low efficiency in its testing phase due to the lack of an explicit classification model. Recently, kNN model-based classifiers have been proposed to improve the conventional kNN. However, their building incurs high computational costs. In this paper, we tackle the model building problem by developing a cluster-based training algorithm to learn an optimized set of representatives that approximate the distributions of training data. The training algorithm adopts a fuzzy clustering method for unsupervised learning on the partial training set, and has a linear time complexity with respect to the size of the set. The experimental results conducted on real-world datasets demonstrate that the new method outperforms the previous kNN model-based classifier in the accuracy and possesses outstanding efficiency compared to kNN based classifiers.

Intelligent Data Analysis, 2000

This paper presents a new approach to unsupervised pattern classification. The classification scheme consists of two main stages. The first one is an unsupervised fuzzy learning procedure, which allows, using a similarity measure and a corresponding threshold, to seek clusters within a set of totally unlabeled samples. It provides, for each detected cluster, a good initial prototype as well as the membership degree of each sample. The second stage is an optimization procedure involving the fuzzy c-means (FCM) algorithm. Both procedures are repeated for different values of the similarity threshold, and three validity criteria are used to assess and rank the quality of all resulting partitions. The effectiveness of this approach is demonstrated, for different parameter values, on both artificial and real test data.

International Journal of Fuzzy Systems, 2019

Shared nearest neighbor (SNN) clustering algorithm is a robust graph-based, efficient clustering method that could handle high-dimensional data. The SNN clustering works well when the data consist of clusters that are of diverse in shapes, densities, and sizes but assignment of the data points lying in the boundary regions of overlapping clusters is not accurate. In order to overcome this problem, we have presented an extension of shared nearest neighbor algorithm that have better capability of handling the data points lying in the boundary regions specifically for overlapping cluster by means of fuzzy concept. Extensive experiments were carried out to compare the proposed approach fuzzy shared nearest neighbor clustering (FSNN) with existing clustering methods K-means, Fuzzy C-means, Density_clust, and Shared Nearest Neighbor. The effectiveness of FSNN is evaluated in benchmark datasets. Experimental results using FSNN method show that it can accurately cluster the data points lying in the overlapping partition and generate compact and well-separated clusters as compared to state-of-the-art clustering algorithm. The results obtained using different clustering methods are validated by standard cluster validation measures.

TURKISH JOURNAL OF ELECTRICAL ENGINEERING & COMPUTER SCIENCES

Day by day huge amounts data are produced, and evaluation of these data becomes more difficult. The data obtained should provide meaningful, correct, and accurate information. Therefore, all data must be separated into clusters correctly, and the right information from these clusters must be obtained. Having the correct clusters depends on the clustering algorithm that is used. There are many clustering algorithms. The density-based methods are very important among the groups of clustering methods, as they can find arbitrary shapes. An advanced model of the density-based spatial clustering of applications with noise (DBSCAN) algorithm, called fuzzy neighborhood DBSCAN Gaussian means (FN-DBSCAN-GM), is offered in this study. The main contribution of FN-DBSCAN-GM is to find the parameters automatically and to divide the data into clusters robustly. The effectiveness of FN-DBSCAN-GM has been demonstrated on overlapping datasets (six artificial and two real-life datasets). The performances of these datasets are compared with the percentage of correct classification and validity index. Our experiments showed that this new algorithm was a preferable and robust algorithm.

International Journal of Engineering Research, 2014

In data mining clustering techniques are used to group together the objects showing similar characteristics within the same cluster and the objects demonstrating different characteristics are grouped into clusters. Clustering approaches can be classified into two categories namely-Hard clustering and Soft clustering. In hard clustering data is divided into clusters in such a way that each data item belongs to a single cluster only while soft clustering also known as fuzzy clustering forms clusters such that data elements can belong to more than one cluster based on their membership levels which indicate the degree to which the data elements belong to the different clusters.

Bulletin of Mathematics, 2012

The current data tends to be more complex than conventional data and need dimension reduction. Dimension reduction is important in cluster analysis and creates a smaller data in volume and has the same analytical results as the original representation. A clustering process needs data reduction to obtain an efficient processing time while clustering and mitigate curse of dimensionality. This paper proposes a model for extracting multidimensional data clustering of health database. We implemented four dimension reduction techniques such as Singular Value Decomposition (SVD), Principal Component Analysis (PCA), Self Organizing Map (SOM) and FastICA. The results show that dimension reductions significantly reduce dimension and shorten processing time and also increased performance of cluster in several health datasets.