Hierarchical Conceptual Clustering Research Articles

Exploratory Analysis of Distributional Data Using the Quantile Method

The quantile method transforms each complex object described by different histogram values to a common number of quantile vectors. This paper retraces the authors’ research, including a principal component analysis, unsupervised feature selection using hierarchical conceptual clustering, and lookup table regression model. The purpose is to show that this research is essentially based on the monotone property of quantile vectors and works cooperatively in the exploratory analysis of the given distributional data.

The Lookup Table Regression Model for Histogram-Valued Symbolic Data

This paper presents the Lookup Table Regression Model (LTRM) for histogram-valued symbolic data. We first transform the given symbolic data to a numerical data table by the quantile method. Then, under the selected response variable, we apply the Monotone Blocks Segmentation (MBS) to the obtained numerical data table. If the selected response variable and some remained explanatory variable(s) organize a monotone structure, the MBS generates a Lookup Table composed of interval values. For a given object, we search the nearest value of an explanatory variable, then the corresponding value of the response variable becomes the estimated value. If the response variable and the explanatory variable(s) are covariate but they follow to a non-monotonic structure, we need to divide the given data into several monotone substructures. For this purpose, we apply the hierarchical conceptual clustering to the given data, and we obtain Multiple Lookup Tables by applying the MBS to each of substructures. We show the usefulness of the proposed method by using an artificial data set and real data sets.

A spatial copula interpolation in a random field with application in air pollution data.

Interpolating a skewed conditional spatial random field with missing data is cumbersome in the absence of Gaussianity assumptions. Copulas can capture different types of joint tail characteristics beyond the Gaussian paradigm. Maintaining spatial homogeneity and continuity around the observed random spatial point is also challenging. Especially when interpolating along a spatial surface, the boundary points also demand focus in forming a neighborhood. As a result, importing the concept of hierarchical clustering on the spatial random field is necessary for developing the copula model with the interface of the Expectation-Maximization algorithm and concurrently utilizing the idea of the Bayesian framework. This article introduces a spatial cluster-based C-vine copula and a modified Gaussian distance kernel to derive a novel spatial probability distribution. To make spatial copula interpolation compatible and efficient, we estimate the parameter by employing different techniques. We apply the proposed spatial interpolation approach to the air pollution of Delhi as a crucial circumstantial study to demonstrate this newly developed novel spatial estimation technique.

Community structure concept for catchment classification: A modularity density-based edge betweenness (MDEB) method

Catchment classification is useful for a variety of purposes in hydrologic, environmental, and ecosystem studies. In the context of classification, the concept of community structure, within the realm of complex networks, is particularly attractive and gaining attention in catchment classification studies. Among the many community structure methods, the edge betweenness (EB) method, which applies a hierarchical clustering concept, is one of the most widely used. The EB method, however, is susceptible to the issue of scale (or size) of the network, essentially due to the modularity function that is used to measure the strength of the community structure. To overcome this limitation, the present study proposes an improvement to the EB method. The proposed method, termed as the Modularity Density-based Edge Betweenness (MDEB) method, uses a modularity density function (or D value) by maximization, to obtain the best split of the network. The effectiveness of the MDEB method is evaluated through its application for catchment classification using streamflow data from two large networks: 218 stations from Australia and 639 stations from the United States (US). For each network, three different scenarios in network sizes are studied: (1) the entire network; (2) smaller network sizes based on 100 random realizations, with each realization having 100 and 300 stations for Australia and the US, respectively; and (3) smaller network sizes based on nine different drainage division regions in Australia and 18 different hydrologic units in the US. The classification outcomes from the MDEB method for these three scenarios are compared with those from the EB method. The results indicate that the MDEB method generally performs better than the EB method, for both Australia and the US. The superiority of the MDEB method is evaluated in terms of the number of communities identified and the number of stations that change communities when different network sizes are considered. The catchment communities are also interpreted in terms of the distance-correlation relationship. The results from the present study offer further evidence as to the usefulness and effectiveness of the community structure concept for catchment classification, especially the proposed MDEB method.

Hierarchical conceptual clustering based on quantile method for identifying microscopic details in distributional data

Symbolic data is aggregated from bigger traditional datasets in order to hide entry specific details and to enable analysing large amounts of data, like big data, which would otherwise not be possible. Symbolic data may appear in many different but complex forms like intervals and histograms. Identifying patterns and finding similarities between objects is one of the most fundamental tasks of data mining. In order to accurately cluster these sophisticated data types, usual methods are not enough. Throughout the years different approaches have been proposed but they mainly concentrate on the “macroscopic” similarities between objects. Distributional data, for example symbolic data, has been aggregated from sets of large data and thus even the smallest microscopic differences and similarities become extremely important. In this paper a method is proposed for clustering distributional data based on these microscopic similarities by using quantile values. Having multiple points for comparison enables to identify similarities in small sections of distribution while producing more adequate hierarchical concepts. Proposed algorithm, called microscopic hierarchical conceptual clustering, has a monotone property and has been found to produce more adequate conceptual clusters during experimentation. Furthermore, thanks to the usage of quantiles, this algorithm allows us to compare different types of symbolic data easily without any additional complexity.

Curse of Dimensionality in Paradoxical High Dimensional Clinical Datasets A Survey

Data storage and retrieval is one among the challenging process in the field of computation. The storage and retrieval of multi-dimensional unstructured conflict data are needs the conversion of structure process. The storage has the major impact on access and computation time. A significant analysis of very large data sets involves different types of datasets as paradoxical high dimensional data. The ideal case assumptions are that data are collected in equal length intervals and while comparing the length are not valid for many real data sets especially clinical data sets. In addition the datasets are different from each other, the data are paradoxical and varies by each medical data. In this paper, the concept of hierarchical clustering with dendrogram structure is used to represent the paradoxical high dimensional clinical datasets. These large clusters of high dimensional datasets are of different dimensions and they may produce much noise and mask the real data to be diverse. There is a survey of clustering techniques used in paradoxical high dimensional clinical datasets and which will be highlighted by the dendrogram representation and also to reduces the dimensions of different clusters.

An Intrusion Detection and Prevention System based on Automatic Learning of Traffic Anomalies

The ever changing network traffic reveals new attack types, which represent a security threat that poses a serious risk for enterprise resources. Therefore, the security administrators are in a real need to employ efficient Intrusion Detection and Prevention Systems, IDPS. Such systems might be capable to learn from the network behavior. In this paper, we present an incremental Learnable Model for Anomaly Detection and Prevention of Zero-day attacks, LMAD/PZ. To facilitate the ability of learning from observations that can provide a reliable model for automatic prevention, a comparison has been carried out between supervised and unsupervised learning techniques. Thus, in LMAD/PZ, the intrusion detection step is integrated with an intrusion prevention plan. To ensure that the prevention plan is dependable and automatic, it must be backed and sustained with robust and accurate detection process. Therefore, two incremental data mining techniques are deeply investigated and implemented on NSL-KDD’99 intrusion dataset. The first technique is the Algorithm Quasi-optimal (AQ), which is a supervised Attributional Rules Learner, ARL, while the second is the Cobweb; an unsupervised hierarchical conceptual clustering algorithm. These algorithms categorize the network connections as either normal or anomalous. The performance of AQ is compared to Cobweb, and the best performance result is integrated with the prevention plan, to afford a fully automated system. The experimental results showed that, the model automatically adapts its knowledge base from continuous network streams, in addition to offering the advantage of detecting novel and zero day attacks. Many experiments have verified that AQ performance outperforms the Cobweb clustering, in terms of accuracy, detection rate and false alarm rate.

Proximity Measurement Technique for Gene Expression Data

Data Mining is an analytical process intended to explore the data in search of consistent patterns. Due to its wide spread applications in biomedical industry and publicly available genomic data, data mining has become upcoming topic in the analysis of gene expression data. Clustering is the first step in understanding the complicated biological systems. The objective of clustering is to organize the samples into intrinsic clusters such that samples with high similarity belong to same cluster. The significance of clustering gene profiles is two-fold. Firstly, it assists in diagnosis of the disease condition and secondly it discloses the effect of certain treatment on genes. In this paper, we propose a new method to cluster gene expression data that is solely based on the concept of hierarchical clustering with a different method to compute the similarity between datasets and merge the pairs. The experimental results on two microarray data show the correctness and competence of proposed technique. Index Terms—Data mining, microarray, gene expression data, hierarchical clustering.

V11. Functional hierarchy within an overall network for visual motion processing and ocular-motor control at rest

Introduction Visual motion processing on one hand and ocular motor functions on the other are rarely studied together in vivo in humans. The interrelation of these functional networks is rather unclear, even though their functional dependence seems obvious. In several fMRI studies the essential nodes of both networks could be localized using voluntary optokinetic ('look') nystagmus (OKN) in the horizontal plane incorporating visual motion tracking (Dieterich et al., 2009). Here, functional connectivity (FC) between these nodes representing both networks was studies using resting-state FC. Methods Resting-state fMRI data of 200 healthy adults (age 44.1±17.9; 79 male) were included in the cross correlation analysis of 9 bilateral nodes including frontal (FEF), supplementary (SEF), cingulate (CEF) and parietal eye fields (PEF), V5 and V6, as well as the superior colliculus (SupCol), the lateral geniculate body (CGL) and the globus pallidus (GlobPal). The necessary ROIs were obtained in a separate OKN fmri experiment with 21 healthy subjects. After spatial preprocessing and confound removal using 24 motion regression and mean signal within white matter and cerebral spinal fluid and band-pass filtering (0.01–0.08Hz), each ROI was represented by the first eigenvariate of the respective voxels' time-series. For each node pair partial correlations were computed within subjects and Fisher Z transformed for one group t -tests corrected for the influence of age and gender resulting in an 18×18 cross correlation matrix. Subsequently, hierarchical cluster analysis was applied to analyze sub-clustering within the overall network. Results The analysis showed consistent FC between each regions respective homotopical partner. Hierarchical clustering revealed an overall split between one cluster comprising SEF, FEF, CEF and GlobPal and a second including CGL, SubCol, V6 as well as PEF and V5. Conclusion This new approach revealed an observer- and task-indepedent separation of the cortical eye fields into two main groups either responsible for voluntary ocular motor control (SEF, FEF, CEF, and GlobPal) or involved in visual motion target tracking streams (CGL, SupCol, PEF, V5, and V6). The further subgrouping of PEF and V5 together with area V6 representing an isolated cluster within the second group seem to reflect their order and importance along the dorsal visual stream known from lesion studies (Pierrot-Deseilligny et al., 2004). The other subgroup of CGL and SupCol may represent the early nodes of the network involved in voluntary and reflexive ocular motor control. In sum, these neuroscientifically sound network findings for visual motion and ocular motor control derived from task-free data give a promising outlook for the novel concept of hierarchical clustering (Figs. 1 and 2).

A Hierarchical Clustering Approach Based on Three-Dimensional Gray Relational Analysis for Clustering a Large Group of Decision Makers with Double Information

Two types of information, collectively referred to as double information, are usually required in management decision-making. The first is preference information expressed in a judgment matrix. The second is reference information expressed in a multi-attribute decision matrix. In this paper, we investigate large-scale group clustering problems with double information in group decision-making. We first establish a novel three-dimensional gray correlation degree index, which integrates the alternative decision-making vector, index vector and alternative preference vector, to fully excavate the correlation between decision makers with double information. We then develop a new clustering procedure combining three-dimensional gray relational analysis and the concept of hierarchical clustering. Moreover, a model for determining clustering centers is established on the basis of the maximum gray correlation degree within each cluster and minimum gray correlation degree among clusters. A heuristic algorithm for the model to identify the core decision maker in each cluster is proposed. Finally, we illustrate the applications of the developed procedures with a practical case. The rationality of the proposed method is demonstrated by comparing results with results obtained using other methods, including the traditional gray clustering method and hierarchical clustering method with single information; i.e., preference information or reference information.

Wormhole Attack Mitigation in Manet: A Cluster Based Avoidance Technique

A Mobile Ad-Hoc Network (MANET) is a self configuring, infrastructure less network of mobile devices connected by wireless links. Loopholes like wireless medium, lack of a fixed infrastructure, dynamic topology, rapid deployment practices, and the hostile environments in which they may be deployed, make MANET vulnerable to a wide range of security attacks and Wormhole attack is one of them. During this attack a malicious node captures packets from one location in the network, and tunnels them to another colluding malicious node at a distant point, which replays them locally. This paper presents a cluster based Wormhole attack avoidance technique. The concept of hierarchical clustering with a novel hierarchical 32bit node addressing scheme is used for avoiding the attacking path during the route discovery phase of the DSR protocol, which is considered as the under lying routing protocol. Pinpointing the location of the wormhole nodes in the case of exposed attack is also given by using this method.

Modified Genetic Algorithm for Flexible Job-Shop Scheduling Problems

This paper proposes a modified version of the genetic algorithm for flexible job-shop scheduling problems (FJSP). The genetic algorithm (GA), a class of stochastic search algorithms, is very effective at finding optimal solutions to a wide variety of problems. The proposed modified GA consists of 1) an effective selection method called “fuzzy roulette wheel selection,” 2) a new crossover operator that uses a hierarchical clustering concept to cluster the population in each generation, and 3) a new mutation operator that helps in maintaining population diversity and overcoming premature convergence. The objective of this research is to find a schedule that minimizes the makespan of the FJSP. The experimental results on 10 well-known benchmark instances show that the proposed algorithm is quite efficient in solving flexible job-shop scheduling problems.

Hierarchical fuzzy clustering decision tree for classifying recipes of ion implanter

In this paper, we propose a hierarchical fuzzy clustering decision tree (HFCDT) for the classification problem with large number of classes and continuous attributes. The HFCDT combines a division-degree matrix based hierarchal clustering technique with the entropy-based C4.5 decision tree algorithm. A hierarchical clustering concept is introduced to achieve a finer fuzzy partition. The hierarchical clustering technique splits the data set into leaf clusters using splitting attributes based on a division-degree matrix and fuzzy rules. The leaf clusters consisting of the data of more than one class will be further classified using the C4.5 algorithm. We have successfully applied the HFCDT for classifying recipes of the working wafers in an ion implanter, and compared the classification results and the training time with the existing software See5 and CART. The comparison results show that the HFCDT not only performs better than See5 and CART in the aspect of 10-fold cross validation for the average of total classification error rates but also consumes less training time. Thus, HFCDT obtains a very successful classification result. This also demonstrates why the hierarchical clustering technique helps reduce the computational complexity of the C4.5 algorithm.

GECO: Galaxy Evolution COde – A new semi-analytical model of galaxy formation

We present a new semi-analytical model of galaxy formation, GECO (Galaxy Evolution COde), aimed at a better understanding of when and how the two processes of star formation and galaxy assembly have taken place. Our model is structured into a Monte Carlo algorithm based on the Extended Press-Schechter theory, for the representation of the merging hierarchy of dark matter halos, and a set of analytic algorithms for the treatment of the baryonic physics, including classical recipes for the gas cooling, the star formation time-scales, galaxy mergers and SN feedback. Together with the galaxies, the parallel growth of BHs is followed in time and their feedback on the hosting galaxies is modelled. We set the model free parameters by matching with data on local stellar mass functions and the BH-bulge relation at z=0. Based on such local boundary conditions, we investigate how data on the high-redshift universe constrain our understanding of the physical processes driving the evolution, focusing in particular on the assembly of stellar mass and on the star formation history. Since both processes are currently strongly constrained by cosmological near- and far-IR surveys, the basic physics of the Lambda CDM hierarchical clustering concept of galaxy formation can be effectively tested by us by comparison with the most reliable set of observables. Our investigation shows that when the time-scales of the stellar formation and mass assembly are studied as a function of dark matter halo mass and the single galaxy stellar mass, the 'downsizing' fashion of star formation appears to be a natural outcome of the model, reproduced even in the absence of the AGN feedback. On the contrary, the stellar mass assembly history turns out to follow a more standard hierarchical pattern progressive in cosmic time, with the more massive systems assembled at late times mainly through dissipationless mergers.

Exploring the wild birds’ migration data for the disease spread study of H5N1: a clustering and association approach

Knowledge about the wetland use of migratory bird species during the annual life circle is very interesting to biologists, as it is critically important in many decision-making processes such as for conservation site construction and avian influenza control. The raw data of the habitat areas and the migration routes are usually in large scale and with high complexity when they are determined by high-tech GPS satellite telemetry. In this paper, we convert these biological problems into computational studies and introduce efficient algorithms for the data analysis. Our key idea is the concept of hierarchical clustering for migration habitat localizations, and the notion of association rules for the discovery of migration routes from the scattered location points in the GIS. One of our clustering results is a tree structure, specially called spatial-tree, which is an illusive map depicting the breeding and wintering home range of bar-headed geese. A related result to this observation is an association pattern that reveals a high possibility that bar-headed geese’s potential autumn migration routes are likely between the breeding sites in the Qinghai Lake, China and the wintering sites in Tibet river valley. Given the susceptibility of geese to spread H5N1, and on the basis of the chronology and the rates of the bar-headed geese migration movements, we can conjecture that bar-headed geese play an important role in the spread of the H5N1 virus at a regional scale in Qinghai-Tibetan Plateau.

ABOUT DATABASE SUMMARIZATION

The summarization system takes a Database (DB) table as input and produces a reduced version of this table through both a rewriting and a generalization process. The resulting table provides records with less precision than the original but it is very informative of the actual DB content. This reduced form can be used as input for advanced Data Mining processes. Several approaches of DB summarization have been proposed in the literature. The most recent is the SaintEtiQ summarization model, proposed initially by Raschia.1 Based on a hierarchical conceptual clustering algorithm, SaintEtiQ builds a summary hierarchy from DB records. In this paper, we propose to extend this DB summarization model by introducing some optimization processes including: (i) minimization of the expert risks domain, (iii) building of the summary hierarchy from DB records, and (iv) cooperation with the user by giving him summaries in different hierarchy levels.

A Methodology for Product Family Ontology Development Using Formal Concept Analysis and Web Ontology Language

The use of ontologies for information sharing is well documented in the literature, but the lack of a comprehensive and systematic methodology for constructing product ontologies has limited the process of developing ontologies for design artifacts. In this paper we introduce the Product Family Ontology Development Methodology (PFODM), a novel methodology to develop formal product ontologies using the Semantic Web paradigm. Within PFODM, Formal Concept Analysis (FCA) is used first to identify similarities among a finite set of design artifacts based on their properties and then to develop and refine a product family ontology using Web Ontology Language (OWL). A family of seven one-time-use cameras is used to demonstrate the steps of the PFODM to construct such an ontology. The benefit of PFODM lies in providing a systematic and consistent methodology for constructing ontologies to support product family design. The resulting ontologies provide a hierarchical conceptual clustering of related design artifacts, which is particularly advantageous for product family design where parts, processes, and most important, information is intentionally shared and reused to reduce complexity, lead-time, and development costs. Potential uses of the resulting ontologies and FCA representations within product family design are also discussed.

S AINTE TIQ: a fuzzy set-based approach to database summarization

In this paper, a new approach to database summarization is introduced through our model named S AINTE TIQ. Based on a hierarchical conceptual clustering algorithm, S AINTE TIQ incrementally builds a summary hierarchy from database records. Furthermore, the fuzzy set-based representation of data allows to handle vague, uncertain or imprecise information, as well as to improve accuracy and robustness of the construction process of summaries. Finally, background knowledge provides a user-defined vocabulary to synthesize and to make highly intelligible the summary descriptions.

GRAPH-BASED HIERARCHICAL CONCEPTUAL CLUSTERING

Hierarchical conceptual clustering has proven to be a useful, although greatly under-explored data mining technique. A graph-based representation of structural information combined with a substructure discovery technique has been shown to be successful in knowledge discovery. The SUBDUE substructure discovery system provides the advantages of both approaches. This work presents SUBDUE and the development of its clustering functionalities. Several examples are used to illustrate the validity of the approach both in structured and unstructured domains, as well as compare SUBDUE to earlier clustering algorithms. Results show that SUBDUE successfully discovers hierarchical clusterings in both structured and unstructured data.

A look at fastfood competition in the Philippines

The study tried to develop a product‐market structure of the fast‐food restaurants in the Philippines. Using the concept of hierarchical clustering based on substitution‐in‐use, the influence of three categories of situational influence were used to modify the original choice‐market mix relationship at every clustering stage. Changes in the importance of the individual components of the product‐service mix served as the basis of the link‐up among competing fast‐food restaurants.