Conventional Clustering Methods Research Articles

A General Multiobjective Clustering Approach Based on Multiple Distance Measures

Many conventional clustering methods have limitations to partition data sets with different structures. The reason is that the relationship of each pair of data objects in different structures is usually based on different distance measures while conventional clustering methods are often designed for an assumption distribution in Euclidean space. Most of current clustering methods have also been proposed for integrating different distance measures together, however, the weights for different distance measures are difficult to set. To alleviate this case and to generate reliable clustering results for data sets with different structures, in this paper, a novel multiple distance measures clustering method based on a multiobjective evolutionary algorithm is proposed to this problem. This approach takes two types of distance measures as multiple objective functions and optimizes them simultaneously by using a modified multiobjective evolutionary algorithm with some new strategies including initialization, crossover operator, mutation operator, and objective functions designing. Moreover, an updated approach was also proposed for detecting the correct cluster number automatically. The new approaches are applied to many datasets with spherical and irregular structures, and the results of eight artificial, four widely used and four real data sets will be exhibited in experiments. The comparisons with other clustering algorithms show that, no matter what shape dataset has, both of the proposed approaches can get satisfactory results in combining different distance measures and detecting the optimal cluster number in a single run.

Application of Grey Clustering Method Based on Improved Analytic Hierarchy Process in Water Quality Evaluation

To highlight the differences in water quality impacts of different indicators in water samples, this paper proposes a grey clustering method based on improved analytic hierarchy process to evaluate the quality of surface water. According to the pollution degree of different indicators in the water quality sample, the importance score is assigned, and the weight of different indicators is calculated by the analytic hierarchy process. The weight participates in the calculation of the grey clustering coefficient, and the evaluated water quality category considers the difference of the contribution rate of different pollutant indicators. The water quality samples of three water periods (Abundant, Normal, and Poor water flow periods) in the four sections of Qingshui River in Duyun City (Tea garden, Youhang, Yingpan and Jiadeng) were selected for evaluation, and the conventional grey clustering method and single factor method were used. The evaluation results were compared and analyzed. The evaluation results show that the improved grey clustering method is more scientific and reasonable and can provide a basis for water quality assessment and water environment management of water environment management departments.

From Partition-Based Clustering to Density-Based Clustering: Fast Find Clusters With Diverse Shapes and Densities in Spatial Databases

Spatial data clustering has played an important role in the knowledge discovery in spatial databases. However, due to the increasing volume and diversity of data, conventional spatial clustering methods are inefficient even on moderately large data sets, and usually fail to discover clusters with diverse shapes and densities. To address these challenges, we propose a two-phase clustering method named KMDD (clustering by combining K-means with density and distance-based method) to fast find clusters with diverse shapes and densities in spatial databases. In the first phase, KMDD uses a partition-based algorithm (K-means) to cluster the data set into several relatively small spherical or ball-shaped subclusters. After that, each subcluster is given a local density; to merge subclusters, KMDD utilizes the idea that genuine cluster cores are characterized by a higher density than their neighbor subclusters and by a relatively large distance from subclusters with higher densities. Extensive experiments on both synthetic and real-world data sets demonstrate that the proposed algorithm has a near-linear time complexity with respect to the data set size and dimension, and has the capability to find clusters with diverse shapes and densities.

Unsupervised Locality-Preserving Robust Latent Low-Rank Recovery-Based Subspace Clustering for Fault Diagnosis

With the increasing demand for unsupervised learning for fault diagnosis, the subspace clustering has been considered as a promising technique enabling unsupervised fault diagnosis. Although various subspace clustering methods have been developed to deal with high-dimensional and non-linear data, analyzing the intrinsic structure from the data is still challenging. To address this issue, a new subspace clustering method based on locality-preserving robust latent low-rank recovery (L2PLRR) was developed. Unlike conventional subspace clustering methods, the developed method maps the high-dimensional and non-linear data into a low-dimensional latent space by preserving local similarities of the data with the goal of resolving the difficulty in analyzing the high-dimensional data. Likewise, in the developed L2PLRR method, learned features correspond to low-rank coefficients of the data in the latent space, which will be further used for fault diagnosis (e.g., identification of health states of an object system). The efficacy of the developed L2PLRR method was verified with a bearing fault diagnosis application by comparing with conventional and state-of-the-art subspace clustering methods in terms of diagnostic performance.

Categorizing software vulnerabilities using overlapping self-organizing map

Software has always been vulnerable to various vulnerably issues. Increasing the number of vulnerabilities and their complexity in the software area has made it more important to categorize them. In this research work, by selecting the MoSCoW prioritization method and by combining it with the SOM self-organizing mapping algorithm, we present a new categorization for the frequent software vulnerabilities. We implemented the proposed method in MATLAB using the relevant tool boxes. The experimental results were evaluated using in-class and out-of-class distance measurements. Classification of software vulnerabilities using OSOM algorithms gives us better results than conventional clustering methods. It can be inferred that the classification of software vulnerabilities is of particular importance in improving the security of a software application. The proposed algorithm can provide an appropriate categorization by taking advantage from the existing overlapping feature.

DEclust: A statistical approach for obtaining differential expression profiles of multiple conditions.

High-throughput RNA sequencing technology is widely used to comprehensively detect and quantify cellular gene expression. Thus, numerous analytical methods have been proposed for identifying differentially expressed genes (DEGs) between paired samples such as tumor and control specimens, but few studies have reported methods for analyzing differential expression under multiple conditions. We propose a novel method, DEclust, for differential expression analysis among more than two matched samples from distinct tissues or conditions. As compared to conventional clustering methods, DEclust more accurately extracts statistically significant gene clusters from multi-conditional transcriptome data, particularly when replicates of quantitative experiments are available. DEclust can be used for any multi-conditional transcriptome data, as well as for extending any DEG detection tool for paired samples to multiple samples. Accordingly, DEclust can be used for a wide range of applications for transcriptome data analysis. DEclust is freely available at http://www.dna.bio.keio.ac.jp/software/DEclust.

Hybrid Gbest-guided Artificial Bee Colony for hard partitional clustering

Clustering is an unsupervised classification method in the field of data mining and plays an essential role in applications in diverse fields. The K-means and K-Medoids are popular examples of conventional partitional clustering methods, which have been prominently applied in various applications. However, they possess several disadvantages, e.g., final solution is dependent on the initial solution, they easily struck into a local optimum solution. The nature-inspired swarm intelligence (SI) methods are global search optimization methods, which offer to be effective to overcome deficiencies of the conventional methods as they possess several desired key features like upgrading the candidate solutions iteratively, decentralization, parallel nature, and a self organizing behavior. The Artificial Bee Colony (ABC) algorithm is one of the recent and well-known SI method, which has been shown effective in various real-world problems. However, it exhibits lack of balance in the exploration and exploitation and shows a poor convergence speed when the number of features (dimensions) increases. Therefore, we make two modifications in it to enhance its exploration and exploitation capabilities to improve quality of the clustering. First, we introduce a gbest-guided search procedure for the fast convergence, which works effectively in large number of features also as it considers all the dimensions simultaneously. Second, in order to avoid being trapped in a local optima and to enhance the information exchange (social learning) between bees for improved search, we incorporate a crossover operator of the genetic algorithm (GA) into it. The proposed strategy is named as Hybrid Gbest-guided Artificial Bee Colony (HGABC) algorithm. We compare clustering results of the HGABC with ABC, variants of the ABC and other recent competitive methods in the swarm and evolutionary intelligence domain on ten real and two synthetic data sets using external quality measures F-measure and Rand-index. The obtained results demonstrate superiority of the proposed method over its competitors in terms of efficiency and effectiveness.

Fast Spectral Clustering With Anchor Graph for Large Hyperspectral Images

The large-scale hyperspectral image (HSI) clustering problem has attracted significant attention in the field of remote sensing. Most traditional graph-based clustering methods still face challenges in the successful application of the large-scale HSI clustering problem mainly due to their high computational complexity. In this letter, we propose a novel approach, called fast spectral clustering with anchor graph (FSCAG), to efficiently deal with the large-scale HSI clustering problem. Specifically, we consider the spectral and spatial properties of HSI in the anchor graph construction. The proposed FSCAG algorithm first constructs anchor graph and then performs spectral analysis on the graph. With this, the computational complexity can be reduced to $O(ndm)$ , which is a significant improvement compared to conventional graph-based clustering methods that need at least $O(n^{2}d)$ , where $n$ , $d$ , and $m$ are the number of samples, features, and anchors, respectively. Several experiments are conducted to demonstrate the efficiency and effectiveness of the proposed FSCAG algorithm.

Revealing community structures by ensemble clustering using group diffusion

We propose an ensemble clustering approach using group diffusion to reveal community structures in data. We represent data points as a directed graph and assume each data point belong to single cluster membership instead of multiple memberships. The method is based on the concept of ensemble group diffusion with a parameter to represent diffusion depth in clustering. The ability to modulate the diffusion-depth parameter by varying it within a certain interval allows for more accurate construction of clusters. Depending on the value of the diffusion-depth parameter, the presented approach can determine very well both local clusters and global structure of data. At the same time, the ability to combine single outcomes of the method results in better cluster segmentation. Due to this property, the proposed method performs well on data sets where other conventional clustering methods fail. We test the method with both simulated and real-world data sets. The results support our theoretical conjectures on improved accuracy compared to other selected methods.

Modified sequential k‐means clustering by utilizing response: A case study for fashion products

AbstractModified sequential k‐means clustering concerns a k‐means clustering problem in which the clustering machine utilizes output similarity in addition. While conventional clustering methods commonly recognize similar instances at features‐level modified sequential clustering takes advantage of response, too. To this end, the approach we pursue is to enhance the quality of clustering by using some proper information. The information enables the clustering machine to detect more patterns and dependencies that may be relevant. This allows one to determine, for instance, which fashion products exhibit similar behaviour in terms of sales. Unfortunately, conventional clustering methods cannot tackle such cases, because they handle attributes solely at the feature level without considering any response. In this study, we introduce a novel approach underlying minimum conditional entropy clustering and show its advantages in terms of data analytics. In particular, we achieve this by modifying the conventional sequential k‐means algorithm. This modified clustering approach has the ability to reflect the response effect in a consistent manner. To verify the feasibility and the performance of this approach, we conducted several experiments based on real data from the apparel industry.

Low Rank Representation on SPD matrices with Log-Euclidean metric

Symmetric Positive semi-Definite (SPD) matrices, as a kind of effective feature descriptors, have been widely used in pattern recognition and computer vision tasks. Affine-invariant metric (AIM) is a popular way to measure the distance between SPD matrices, but it imposes a high computational burden in practice. Compared with AIM, the Log-Euclidean metric embeds the SPD manifold via the matrix logarithm into a Euclidean space in which only classical Euclidean computation is involved. The advantage of using this metric for the non-linear SPD matrices representation of data has been recognized in some domains such as compressed sensing, however one pays little attention to this metric in data clustering. In this paper, we propose a novel Low Rank Representation (LRR) model on SPD matrices space with Log-Euclidean metric (LogELRR), which enables us to handle non-linear data through a linear manipulation manner. To further explore the intrinsic geometry distance between SPD matrices, we embed the SPD matrices into Reproductive Kernel Hilbert Space (RKHS) to form a family of kernels on SPD matrices based on the Log-Euclidean metric and construct a novel kernelized LogELRR method. The clustering results on a wide range of datasets, including object images, facial images, 3D objects, texture images and medical images, show that our proposed methods overcome other conventional clustering methods.

Identification of relevant subtypes via preweighted sparse clustering

Cluster analysis methods are used to identify homogeneous subgroups in a data set. In biomedical applications, one frequently applies cluster analysis in order to identify biologically interesting subgroups. In particular, one may wish to identify subgroups that are associated with a particular outcome of interest. Conventional clustering methods generally do not identify such subgroups, particularly when there are a large number of high-variance features in the data set. Conventional methods may identify clusters associated with these high-variance features when one wishes to obtain secondary clusters that are more interesting biologically or more strongly associated with a particular outcome of interest. A modification of sparse clustering can be used to identify such secondary clusters or clusters associated with an outcome of interest. This method correctly identifies such clusters of interest in several simulation scenarios. The method is also applied to a large prospective cohort study of temporomandibular disorders and a leukemia microarray data set.

Divide‐conquer method for improving possibilistic c‐means

Possibilistic c-means (PCM) was proposed to overcome the problem of the noise sensitivity of fuzzy c-means, but its performance highly depends on the initialisation of cluster centres and often is degraded due to producing coincident clusters or close centres. To tackle these defects of PCM, a divide-conquer method which not only provides appropriate cluster centres but also yields pre-clustered and un-clustered data information which are used to overcome the coincident or close clustering problem is presented. Experiment results on a simulated magnetic resonance brain image data corrupted by noise and bias-field shows that the proposed method has a better clustering performance than conventional PCM clustering methods.

METHODS FOR CLUSTERING TIME SERIES DATA ACQUIRED FROM MOBILE HEALTH APPS.

In our recent Asthma Mobile Health Study (AMHS), thousands of asthma patients across the country contributed medical data through the iPhone Asthma Health App on a daily basis for an extended period of time. The collected data included daily self-reported asthma symptoms, symptom triggers, and real time geographic location information. The AMHS is just one of many studies occurring in the context of now many thousands of mobile health apps aimed at improving wellness and better managing chronic disease conditions, leveraging the passive and active collection of data from mobile, handheld smart devices. The ability to identify patient groups or patterns of symptoms that might predict adverse outcomes such as asthma exacerbations or hospitalizations from these types of large, prospectively collected data sets, would be of significant general interest. However, conventional clustering methods cannot be applied to these types of longitudinally collected data, especially survey data actively collected from app users, given heterogeneous patterns of missing values due to: 1) varying survey response rates among different users, 2) varying survey response rates over time of each user, and 3) non-overlapping periods of enrollment among different users. To handle such complicated missing data structure, we proposed a probability imputation model to infer missing data. We also employed a consensus clustering strategy in tandem with the multiple imputation procedure. Through simulation studies under a range of scenarios reflecting real data conditions, we identified favorable performance of the proposed method over other strategies that impute the missing value through low-rank matrix completion. When applying the proposed new method to study asthma triggers and symptoms collected as part of the AMHS, we identified several patient groups with distinct phenotype patterns. Further validation of the methods described in this paper might be used to identify clinically important patterns in large data sets with complicated missing data structure, improving the ability to use such data sets to identify at-risk populations for potential intervention.

Improved pan-specific prediction of MHC class I peptide binding using a novel receptor clustering data partitioning strategy.

Pan-specific prediction of receptor-ligand interaction is conventionally done using machine-learning methods that integrates information about both receptor and ligand primary sequences. To achieve optimal performance using machine learning, dealing with overfitting and data redundancy is critical. Most often so-called ligand clustering methods have been used to deal with these issues in the context of pan-specific receptor-ligand predictions, and the MHC system the approach has proven highly effective for extrapolating information from a limited set of receptors with well characterized binding motifs, to others with no or very limited experimental characterization. The success of this approach has however proven to depend strongly on the similarity of the query molecule to the molecules with characterized specificity using in the machine-learning process. Here, we outline an alternative strategy with the aim of altering this and construct data sets optimal for training of pan-specific receptor-ligand predictions focusing on receptor similarity rather than ligand similarity. We show that this receptor clustering method consistently in benchmarks covering affinity predictions, MHC ligand and MHC epitope identification perform better than the conventional ligand clustering method on the alleles with remote similarity to the training set.

Supervised Regression Clustering

Clustering techniques typically group similar instances underlying individual attributes by supposing that similar instances have similar attributes characteristic. On contrary, clustering similar instances given a specific behavior is framed through supervised learning. For instance, which fashion products have similar behavior in term of sales. Unfortunately, conventional clustering methods cannot tackle this case, since they handle attributes by a same manner. In fact, conventional clustering approaches do not consider any response, and moreover they assume attributes act by the same importance. However, clustering instances with respect to responses leads to a better data analytics. In this research, the authors introduce an approach for the goal supervised clustering and show its advantage in terms of data analytics as well as prediction. To verify the feasibility and the performance of this approach the authors conducted several experiments on a real dataset derived from an apparel industry.

A Granular Classifier By Means of Context-based Similarity Clustering

In this study, we propose a granular classifier (GC) with the aid of a context-based similarity clustering (CSC) method and applied it for network intrusion detection. The proposed CSC supporting the design of information granules is exploited here to determine the so-called contexts. Unlike the conventional similar clustering method, here the CSC built clusters by taking into consideration of both input data and output data. The design of granular classifier is realized based on the if-then rules, which consists two parts: namely premise part and conclusion part. The premise part is developed by using the CSC, while the conclusion part is realized with the aid of supported vector machines. In contrast to typical rule-based classifier, the underlying principle exploited here is to consider a robust classification with the adequate use of output data. In particular, rule-based classifiers or supported vector machines can be regarded as a special case of the proposed granular classifier. Numeric studies show the superiority of the proposed approach.

Using a Method Based on a Modified K-Means Clustering and Mean Shift Segmentation to Reduce File Sizes and Detect Brain Tumors from Magnetic Resonance (MRI) Images

In this paper, we propose a method of elaborating and detecting brain tumor from MRI suitable for information sharing via the internet for a healthcare provider. This method allows for reducing image sizes without reducing the information content of the images in terms of detecting tumors. The proposed method involves first clarifying the brain tumor area using a modified K-means clustering method and initial segmentation using mean shift segmentation. Then a threshold setting is used to convert the gray scale image and remove noise by applying an erode operation. Finally, the brain tumors in the images are detected using a watershed algorithm. The proposed method was compared with two well-known methods namely the conventional K-mean clustering and Fuzzy C Means (FCM) clustering. We verified the precision and the objectivity of our proposed method. The average precision and recall for our proposed method were excellent with values of 0.914052 and 0.995641, respectively. Our method detected more brain tumors than the conventional K-means clustering and FCM clustering methods and was able to provide for an efficient image data processing with reduced file sizes.

A new eigenvector selection strategy applied to develop spectral clustering

Spectral methods are strong tools that can be used for extraction of the data’s structure based on eigenvectors of constructed affinity matrices. In this paper, we aim to propose some new measurement functions to evaluate the ability of each eigenvector of affinity matrix in data clustering. In the proposed strategy, each eigenvector’s elements are clustered by traditional fuzzy c-means algorithm and then informative eigenvectors selection is performed by optimization of an objective function which defined based on three criterions. These criterions are the compactness of clusters, distance between clusters and stability of clustering to evaluate each eigenvector based on considering the structure of clusters which placed on. Finally, Lagrange multipliers method is used to minimize the proposed objective function and extract the most informative eigenvectors. To indicate the merits of our algorithm, we consider UCI Machine Learning Repository databases, COIL20, YALE-B and PicasaWeb as benchmark data sets. Our simulation’s results confirm the superior performance of the proposed strategy in developing spectral clustering compared to conventional clustering methods and recent eigenvector selection based algorithms.

An improved ant algorithm with LDA-based representation for text document clustering

Document clustering can be applied in document organisation and browsing, document summarisation and classification. The identification of an appropriate representation for textual documents is extremely important for the performance of clustering or classification algorithms. Textual documents suffer from the high dimensionality and irrelevancy of text features. Besides, conventional clustering algorithms suffer from several shortcomings, such as slow convergence and sensitivity to the initial value. To tackle the problems of conventional clustering algorithms, metaheuristic algorithms are frequently applied to clustering. In this paper, an improved ant clustering algorithm is presented, where two novel heuristic methods are proposed to enhance the clustering quality of ant-based clustering. In addition, the latent Dirichlet allocation (LDA) is used to represent textual documents in a compact and efficient way. The clustering quality of the proposed ant clustering algorithm is compared to the conventional clustering algorithms using 25 text benchmarks in terms of F-measure values. The experimental results indicate that the proposed clustering scheme outperforms the compared conventional and metaheuristic clustering methods for textual documents.