Partitional Clustering Algorithms Research Articles

Improved fast partitional clustering algorithm for text clustering

Document clustering has become an important task for processing the big amount of textual information available on the Internet. On the other hand, k-means is the most widely used algorithm for clustering, mainly due to its simplicity and effectiveness. However, k-means becomes slow for large and high dimensional datasets, such as document collections. Recently the FPAC algorithm was proposed to mitigate this problem, but the improvement in the speed was reached at the cost of reducing the quality of the clustering results. For this reason, in this paper, we introduce an improved FPAC algorithm, which, according our experiments on different document collections, allows obtaining better clustering results than FPAC, without highly increasing the runtime.

Unsupervised KPIs-Based Clustering of Jobs in HPC Data Centers.

Performance analysis is an essential task in high-performance computing (HPC) systems, and it is applied for different purposes, such as anomaly detection, optimal resource allocation, and budget planning. HPC monitoring tasks generate a huge number of key performance indicators (KPIs) to supervise the status of the jobs running in these systems. KPIs give data about CPU usage, memory usage, network (interface) traffic, or other sensors that monitor the hardware. Analyzing this data, it is possible to obtain insightful information about running jobs, such as their characteristics, performance, and failures. The main contribution in this paper was to identify which metric/s (KPIs) is/are the most appropriate to identify/classify different types of jobs according to their behavior in the HPC system. With this aim, we had applied different clustering techniques (partition and hierarchical clustering algorithms) using a real dataset from the Galician computation center (CESGA). We concluded that (i) those metrics (KPIs) related to the network (interface) traffic monitoring provided the best cohesion and separation to cluster HPC jobs, and (ii) hierarchical clustering algorithms were the most suitable for this task. Our approach was validated using a different real dataset from the same HPC center.

An Improved Mean Imputation Clustering Algorithm for Incomplete Data

There are many incomplete data sets in all fields of scientific studies due to random noise, data lost, limitations of data acquisition, data misunderstanding etc. Most of the clustering algorithms can not be used for incomplete data sets directly because objects with missing values need to be preprocessed. For this reason, this paper presents an improved mean imputation clustering algorithm for incomplete data based on partition clustering algorithm. In the proposed method, we divide the universe into two sets: the set of objects with non-missing values and the set of objects with missing values. Firstly, the objects with non-missing values are clustered by traditional clustering algorithm. For each object with missing values, we use the mean attribute’s value of each cluster to fill the missing attribute’s value based on the cluster results of the objects with non-missing values, respectively. Perturbation analysis of cluster centroid is applied to search the optimal imputation. The experimental clustering results on some UCI data sets are evaluated by several validity indexes, which proves the effectiveness of the proposed algorithm.

Hybrid of Hierarchical and Partitional Clustering Algorithm for Gene Expression Data

Microarray analysis able to monitor thousands of gene expression data, however, to elucidate the hidden patterns in the data is a complex process. These gene expression data show its imprecision, noise and vagueness due to its high dimensional properties. There are a handful of clustering algorithms have been proposed to extract the important information from the gene expression data. However, identifying the underlying biological knowledge of the data is still hard. To acknowledge these issues, clustering algorithms are used to reduce the data complexity. In this article, hybrid of agglomerative hierarchical clustering and modified k-medoids (partitional clustering) are proposed. Application of the proposed of clustering algorithms to group the genes that have similar functionality which might assist pre-processing procedures. In order to emphasize the quality of the clustering results, cluster quality index (CQI) is determined. Lung and ovary data sets used and the method retrieved a fair clustering with CQI, 0.37 and 0.48 respectively. This research contributes by avoiding biasness toward genes and provide true sense of clustering output using the advantage of hierarchical and partitional clustering methods.

K-PbC: an improved cluster center initialization for categorical data clustering

The performance of a partitional clustering algorithm is influenced by the initial random choice of cluster centers. Different runs of the clustering algorithm on the same data set often yield different results. This paper addresses that challenge by proposing an algorithm named k-PbC, which takes advantage of non-random initialization from the view of pattern mining to improve clustering quality. Specifically, k-PbC first performs a maximal frequent itemset mining approach to find a set of initial clusters. It then uses a kernel-based method to form cluster centers and an information-theoretic based dissimilarity measure to estimate the distance between cluster centers and data objects. An extensive experimental study was performed on various real categorical data sets to draw a comparison between k-PbC and state-of-the-art categorical clustering algorithms in terms of clustering quality. Comparative results have revealed that the proposed initialization method can enhance clustering results and k-PbC outperforms compared algorithms for both internal and external validation metrics.

A constrained agglomerative clustering approach for unipartite and bipartite networks with application to credit networks

Researchers and practitioners have been interested in solving real-world problems through clustering. The clustering of nodes in networks with unipartite or bipartite structure is important to explore real-world complex networks present in nature and society. Bipartite networks form an important class of complex networks because they reveal the heterogeneity of nodes in a network. However, most extant clustering methods focus only on unipartite networks. In this work, a novel constrained agglomerative clustering method applicable to unipartite and bipartite networks has been proposed. Initially, the topology of a network is modeled according to set-theoretic principles. Subsequently, the concepts related to rough set theory and relative linkage are used to cluster the set of nodes. The utility and effectiveness of the proposed approach are demonstrated through offline experiments on unipartite and bipartite networks. A comparison against ten state-of-the-art similarity measures over two different partitional clustering algorithms reveals the effectiveness of the proposed relative linkage measure. Moreover, a comparative analysis with state-of-the-art network clustering methods reveals the viability of the proposed rough set-based constrained agglomerative clustering algorithm. Finally, the proposed method has been applied for the detection of cohesive subgroups of banks in a real bipartite network formed by mapping credit relationships between Indian firms and banks.

A Framework to Perform Asset Allocation Based on Partitional Clustering

Over the past years, many approaches to perform asset allocation have been proposed in the literature. Most of them tackle this problem as an optimization task, where the goal is to maximize return, whilst minimizing the risk. However, such approaches require the inversion of a positive-definite covariance matrix, usually resulting in the concentration of allocation, instability and low performance. Some methods have been recently introduced to solve this problem by facing it as a clustering problem. This paper introduces a framework for asset allocation based on partitional clustering algorithms. The idea is to segment the assets into clusters of correlated assets, allocate resources for each cluster and then within each cluster. The framework allows the use of different partitional clustering algorithms, intragroup and intergroup allocation methods. Also, various assessment criteria are considered, and a specialized initialization method is proposed for the clustering algorithm. The framework is evaluated with the Brazilian Stock Exchange (B3) data from the period 12/2005 to 04/2020. Different initialization methods are used for the clustering algorithm together with two intergroup and two intragroup techniques, resulting in five experimental scenarios. The results are compared with the Ibovespa index, the mean-variance model of Markowitz, and the risk-parity model recently proposed by Lopez de Prado.

Stochastic Numerical P Systems With Application in Data Clustering Problems

This paper proposes an extension of numerical P (NP) systems, called stochastic numerical P (StNP) systems. In StNP systems, a novel stochastic production function-repartition protocol is developed to evolve the variables. Compared with usual production function-repartition protocol in NP systems, there are two differences in stochastic production function-repartition protocol: (i) some stochastic/randomized factors are introduced in production function and/or repartition protocol; (ii) data updating way is relaxed and can be flexibly used. An StNP system has a nested structure and contains some variables and a group of programs. The programs consist of stochastic production function-repartition protocols and/or usual production function-repartition protocols. Therefore, StNP systems are a class of distributed parallel computing models with stochastic and dynamic characteristics. Data clustering problems are used as an application to demonstrate the availability and effectiveness of StNP systems. Based on StNP systems, a novel partition clustering algorithm is presented. Experimental results demonstrate advantage of StNP systems for data clustering problems.

Clustering Mixed Numeric and Categorical Data With Cuckoo Search

Clustering analysis, as an important technique in data mining, aims to identify the nature groups or clusters of data objects in the attribute space. Data objects in real-world applications are commonly described by both numeric and categorical attributes. In this research, considering that the partitional clustering algorithms designed for this type of mixed data are prone to get trapped into local optima and the cuckoo search approach is efficient in solving global optimization problems, we propose CCS-K-Prototypes, a novel partitional Clustering algorithm based on Cuckoo Search and K-Prototypes, for clustering mixed numeric and categorical data. To deal with different types of attributes, we develop a novel representation for candidate solutions, and suggest two formulas for the cuckoo to search for the potential solution around the existing solutions or in the entire attribute space. Finally, the performance of the proposed algorithm is assessed by a series of experiments on five benchmark datasets.

Sequential dimension reduction and clustering of mixed-type data

Clustering of a set of objects described by a mixture of continuous and categorical variables can be a challenging task. In the context of data reduction, an effective class of methods combine dimension reduction with clustering in the reduced space. In this paper, we review three approaches for sequential dimension reduction and clustering of mixed-type data. The first step of each approach involves the application of principal component analysis on a suitably transformed matrix. In the second step, a partitioning or hierarchical clustering algorithm is applied to the object scores in the reduced space. The common theoretical underpinnings of the three approaches are highlighted. The results of a benchmarking study show that sequential dimension reduction and clustering is an effective strategy, especially when categorical variables are more informative than continuous with regard to the underlying cluster structure. Strengths and limitations are also demonstrated on a real mixed-type dataset.

Repartitioned Optimized K-Mean Centroid Based Partitioned Clustering using MapReduce in Analyzing High Dimensional Big Data

With the advent of IoT, number of IOT-devices are deployed in the city to acquisition data. These devices acquire enormous data and to analyze such data one need to configure novel hardware to scale up the existing servers and need to develop an application with précised framework. This work recommends an adapted scale out approach in which huge multi-dimensional datasets can be processed using existing commodity hardware. In this approach, Hadoop Distributed File System (HDFS) holds the huge multi-dimensional data to be processed and it can be processed and analyzed by using MapReduce (MR) framework. In the proposed approach, we implemented an optimized repartitioned K-Means centroid based partitioning clustering algorithm using MR framework for Smart City dataset. This dataset contains 10 million objects and each object has six attributes. The results show that the proposed approach is a scalable approach to compute intra cluster density and inter cluster density effectively.

Structure and kinematics of the Taurus star-forming region from Gaia-DR2 and VLBI astrometry

Aims. We take advantage of the second data release of the Gaia space mission and the state-of-the-art astrometry delivered from very long baseline interferometry observations to revisit the structure and kinematics of the nearby Taurus star-forming region. Methods. We apply a hierarchical clustering algorithm for partitioning the stars in our sample into groups (i.e., clusters) that are associated with the various molecular clouds of the complex, and derive the distance and spatial velocity of individual stars and their corresponding molecular clouds. Results. We show that the molecular clouds are located at different distances and confirm the existence of important depth effects in this region reported in previous studies. For example, we find that the L 1495 molecular cloud is located at d = 129.9+0.4−0.3 pc, while the filamentary structure connected to it (in the plane of the sky) is at d = 160.0+1.2−1.2 pc. We report B 215 and L 1558 as the closest (d = 128.5+1.6−1.6 pc) and most remote (d = 198.1+2.5−2.5 pc) substructures of the complex, respectively. The median inter-cloud distance is 25 pc and the relative motion of the subgroups is on the order of a few km s−1. We find no clear evidence for expansion (or contraction) of the Taurus complex, but signs of the potential effects of a global rotation. Finally, we compare the radial velocity of the stars with the velocity of the underlying 13CO molecular gas and report a mean difference of 0.04 ± 0.12 km s−1 (with rms of 0.63 km s−1) confirming that the stars and the gas are tightly coupled.

How many models/atlases are needed as priors for capturing anatomic population variations?

Many medical image processing and analysis operations can benefit a great deal from prior information encoded in the form of models/atlases to capture variations over a population in form, shape, anatomic layout, and image appearance of objects. However, two fundamental questions have not been addressed in the literature: "How many models/atlases are needed for optimally encoding prior information to address the differing body habitus factor in that population?" and "Images of how many subjects in the given population are needed to optimally harness prior information?" We propose a method to seek answers to these questions. We assume that there is a well-defined body region of interest and a subject population under consideration, and that we are given a set of representative images of the body region for the population. After images are trimmed to the exact body region, a hierarchical agglomerative clustering algorithm partitions the set of images into a specified number of groups by using pairwise image (dis)similarity as a cost function. Optionally the images may be pre-registered among themselves prior to this partitioning operation. We define a measure called Residual Dissimilarity (RD) to determine the goodness of each partition. We then ascertain how RD varies as a function of the number of elements in the partition for finding the optimum number(s) of groups. Breakpoints in this function are taken as the recommended number of groups/models/atlases. Our results from analysis of sizeable CT data sets of adult patients from two body regions - thorax (346) and head and neck (298) - can be summarized as follows. (1) A minimum of 5 to 8 groups (or models/atlases) seems essential to properly capture information about differing anatomic forms and body habitus. (2) A minimum of 150 images from different subjects in a population seems essential to cover the anatomical variations for a given body region. (3) In grouping, body habitus variations seem to override differences due to other factors such as gender, with/without contrast enhancement in image acquisition, and presence of moderate pathology. This method may be helpful for constructing high quality models/atlases from a sufficiently large population of images and in optimally selecting the training image sets needed in deep learning strategies.

Two-Stage Load Profiling of HV Feeders of a Distribution System

This paper describes a two-stage unsupervised clustering based on the g-Means approach, which can employ any partitional or crisp clustering algorithm for load profiling of high-voltage (HV) feeders of a distribution system. The methodology proposed can evaluate number of clusters and its respective members from input data. The outcome of g-Means is dependent on a significance level— $\alpha$ . This paper is based on feeder loading data, which are recorded at a transformer secondary at a distribution substation. Real-life active power data of 33-kV feeders acquired through existing supervisory control and data acquisition (SCADA) system of Central Indian state distribution company have been taken as a case study, as it is being processed for the first time, and hence, no prima facie information regarding the number of clusters or classes was available. The first stage entails checking for the randomness in daily load patterns and to extract month-wise working day (weekday) and nonworking day (holidays/Sundays) patterns. The second stage is based on this outcome, where clustering is performed to group similar patterns into clusters to generate month-wise “typical monthly load profiles”. The results are acceptable and are validated by using appropriate validity indices.

Fuzzy clustering and fuzzy c-means partition cluster analysis and validation studies on a subset of citescore dataset

A hard partition clustering algorithm assigns equally distant points to one of the clusters, where each datum has the probability to appear in simultaneous assignment to further clusters. The fuzzy cluster analysis assigns membership coefficients of data points which are equidistant between two clusters so the information directs have a place toward in excess of one cluster in the meantime. For a subset of CiteScore dataset, fuzzy clustering (fanny) and fuzzy c-means (fcm) algorithms were implemented to study the data points that lie equally distant from each other. Before analysis, clusterability of the dataset was evaluated with Hopkins statistic which resulted in 0.4371, a value < 0.5, indicating that the data is highly clusterable. The optimal clusters were determined using NbClust package, where it is evidenced that 9 various indices proposed 3 cluster solutions as best clusters. Further, appropriate value of fuzziness parameter <em>m</em> was evaluated to determine the distribution of membership values with variation in <em>m</em> from 1 to 2. Coefficient of variation (CV), also known as relative variability was evaluated to study the spread of data. The time complexity of fuzzy clustering (fanny) and fuzzy c-means algorithms were evaluated by keeping data points constant and varying number of clusters.

Параллельный алгоритм кластеризации данных для многоядерных ускорителей Intel MIC

Алгоритм PAM (Partitioning Around Medoids) представляет собой разделительный алгоритм кластеризации, в котором в качестве центров кластеров выбираются только кластеризуемые объекты (медоиды). Кластеризация на основе техники медоидов применяется в широком спектре приложений: сегментирование медицинских и спутниковых изображений, анализ ДНК-микрочипов и текстов и др. На сегодня имеются параллельные реализации PAM для систем GPU и FPGA, но отсутствуют таковые для многоядерных ускорителей архитектуры Intel Many Integrated Core (MIC). В настоящей статье предлагается новый параллельный алгоритм кластеризации PhiPAM для ускорителей Intel MIC. Вычисления распараллеливаются с помощью технологии OpenMP. Алгоритм предполагает использование специализированной компоновки данных в памяти и техники тайлинга, позволяющих эффективно векторизовать вычисления на системах Intel MIC. Эксперименты, проведенные на реальных наборах данных, показали хорошую масштабируемость алгоритма. The PAM (Partitioning Around Medoids) is a partitioning clustering algorithm where each cluster is represented by an object from the input dataset (called a medoid). The medoid-based clustering is used in a wide range of applications: the segmentation of medical and satellite images, the analysis of DNA microarrays and texts, etc. Currently, there are parallel implementations of PAM for GPU and FPGA systems, but not for Intel Many Integrated Core (MIC) accelerators. In this paper, we propose a novel parallel PhiPAM clustering algorithm for Intel MIC systems. Computations are parallelized by the OpenMP technology. The algorithm exploits a sophisticated memory data layout and loop tiling technique, which allows one to efficiently vectorize computations with Intel MIC. Experiments performed on real data sets show a good scalability of the algorithm.

An improved approach to fuzzy clustering based on FCM algorithm and extended VIKOR method

Fuzzy C-means algorithm is a fuzzy partitional clustering algorithm. However, accuracy and easy to implement have converted this algorithm to the focus of research, and sensitivity to noisy data is an important and challenging issue in the algorithm, so that in recent years, many studies have been done to improve it. In this paper, a clustering algorithm named Fuzzy VIKOR C-means presented that by utilizing the extended VIKOR method based on targeted displacements in the centroids of the clusters seek to benefit from the flexibility property. Moreover, this algorithm also, considering Dunn’s index, means, and density measures as profit criteria, and DB index and the entropy measures as cost criteria, can reduce the sensitivity to noisy data and can enhance performance and quality of clusters. According to the simulation results and comparison with some recent well-known methods, this approach has an effective role in improving the assessment criteria.

A Fast Multiobjective Fuzzy Clustering with Multimeasures Combination

Most of the existing clustering algorithms are often based on Euclidean distance measure. However, only using Euclidean distance measure may not be sufficient enough to partition a dataset with different structures. Thus, it is necessary to combine multiple distance measures into clustering. However, the weights for different distance measures are hard to set. Accordingly, it appears natural to keep multiple distance measures separately and to optimize them simultaneously by applying a multiobjective optimization technique. Recently a new clustering algorithm called ‘multiobjective evolutionary clustering based on combining multiple distance measures’ (MOECDM) was proposed to integrate Euclidean and Path distance measures together for partitioning the dataset with different structures. However, it is time-consuming due to the large-sized genes. This paper proposes a fast multiobjective fuzzy clustering algorithm for partitioning the dataset with different structures. In this algorithm, a real encoding scheme is adopted to represent the individual. Two fuzzy clustering objective functions are designed based on Euclidean and Path distance measures, respectively, to evaluate the goodness of each individual. An improved evolutionary operator is also introduced accordingly to increase the convergence speed and the diversity of the population. In the final generation, a set of nondominated solutions can be obtained. The best solution and the best distance measure are selected by using a semisupervised method. Afterwards, an updated algorithm is also designed to detect the optimal cluster number automatically. The proposed algorithms are applied to many datasets with different structures, and the results of eight artificial and six real-life datasets are shown in experiments. Experimental results have shown that the proposed algorithms can not only successfully partition the dataset with different structures, but also reduce the computational cost.

Selection of 'K' in K-means clustering using GA and VMA

The K-means algorithm is the most widely used partitional clustering algorithms. In spite of several advances in K-means clustering algorithm, it suffers in some drawbacks like initial cluster centres, stuck in local optima etc. The initial guessing of cluster centres lead to the bad clustering results in K-means and this is one of the major drawbacks of K-means algorithm. In this paper, a new strategy is proposed where we have blended K-means algorithm with genetic algorithm (GA) and volume metric algorithm (VMA) to predict the best value of initial cluster centres, which is not in the case of only K-means algorithm. The paper concludes with the analysis of the results of using the proposed measure to determine the number of clusters for the K-means algorithm for different well-known datasets from UCI machine learning repository.

Using bagging to enhance clustering procedures for planar shapes

Partitional clustering algorithms find a partition maximising or minimising some numerical criterion. Statistical shape analysis is used to make decisions observing the shape of objects. The shape of an object is the remaining information when the effects of location, scale and rotation are removed. This paper introduces clustering algorithms suitable for planar shapes. Four numerical criteria are adapted to each algorithm. In order to escape from local optima to reach a better clustering, these algorithms are performed in the framework of bagging procedures. Simulation studies are carried to validate these proposed methods and two real-life datasets are also considered. The experiment quality is assessed by the corrected Rand index and the results the application of the proposed algorithms showed the effectiveness of these algorithms using different clustering criteria and the union of the bagging method to the cluster algorithms provided substantial gains in of the quality of the clusters.