Showing papers on "Cluster analysis published in 1973"

Pattern classification and scene analysis

Abstract: Provides a unified, comprehensive and up-to-date treatment of both statistical and descriptive methods for pattern recognition. The topics treated include Bayesian decision theory, supervised and unsupervised learning, nonparametric techniques, discriminant analysis, clustering, preprosessing of pictorial data, spatial filtering, shape description techniques, perspective transformations, projective invariants, linguistic procedures, and artificial intelligence techniques for scene analysis.

Clustering Using a Similarity Measure Based on Shared Near Neighbors

Abstract: A nonparametric clustering technique incorporating the concept of similarity based on the sharing of near neighbors is presented. In addition to being an essentially paraliel approach, the computational elegance of the method is such that the scheme is applicable to a wide class of practical problems involving large sample size and high dimensionality. No attempt is made to show how a priori problem knowledge can be introduced into the procedure.

A computer generated aid for cluster analysis

Abstract: A computer generated graphic method, which can be used in conjunction with any hierarchical scheme of cluster analysis, is described and illustrated. The graphic principle used is the representation of the elements of a data matrix of similarities or dissimilarities by computer printed symbols (of character overstrikes) of various shades of darkness, where a dark symbol corresponds to a small dissimilarity. The plots, applied to a data matrix before clustering and to the rearranged matrix after clustering, show at a glance whether clustering brought forth any distinctive clusters. A well-known set of data consisting of the correlations of 24 psychological tests is used to illustrate the comparison of groupings by four methods of factor analysis and two methods of cluster analysis.

Min and max hierarchical clustering using asymmetric similarity measures

Abstract: The min and the max hierarchical clustering methods discussed by Johnson are extended to include the use of asymmetric similarity values. The first part of the paper presents the basic min and max procedures but in the context of graph theory; this description is then generalized to directed graphs as a way of introducing the less restrictive characterization of the original clustering techniques.

ClusteringUsing a Similarity Measure Based on SharedNear Neighbors

Abstract: A nonparametric clustering technique incorporating the concept ofsimilarity basedonthesharing ofnearneighbors ispre- sented. Inaddition tobeing anessentially paraliel approach, thecom- putational elegance ofthemethodissuchthattheschemeisapplicable toawideclass ofpractical problems involving large sample size andhigh dimensionality. No attempt ismadetoshowhowapriori problem knowledge canbeintroduced into theprocedure. IndexTerms-Clustering, nonparametric, pattern recognition, shared nearneighbors, similarity measure.

Monotone invariant clustering procedures

Abstract: A major justification for the hierarchical clustering methods proposed by Johnson is based upon their invariance with respect to monotone increasing transformations of the original similarity measures Several alternative procedures are presented in this paper that also share in the same property of invariance One of these techniques constructs a hierarchy of partitions by sequentially minimizing a monotone invariant goodness-of-fit statistic; the other techniques construct a hierarchy of partitions by successively subdividing the complete set of objects until one partition class is defined for each individual member in the set A numerical example comparing these alternative procedures with Johnson's two methods is duscussed in terms of a simplified computational scheme for obtaining the necessary hierarchies

The Dynamic Clusters Method in Nonhierarchical Clustering

Abstract: Given a finite setE ⊂Rn, the problem is to find clusters (or subsets of “similar” points inE) and at the same time to find the most typical elements of this set. An original mathematical formulation is given to the problem. The proposed algorithm operates on groups of points, called “samplings” (“samplings” may be called “multiple centers” or “cores”); these “samplings” adapt and evolve into interesting clusters. Compared with other clustering algorithms, this algorithm requires less machine time and storage. We provide some propositions about nonprobabilistic convergence and a sufficient condition which ensures the decrease of the criterion. Some computational experiments are presented.

Population mixture models and clustering algorithms

Abstract: The problem of clustering individuals is considered within the context of a mixture of distributions. A modification of the usual approach to population mixtures is employed. As usual, a parametric family of distributions is considered, a set of parameter values being associated with each population. In addition, with each observation is associated an identification parameter, Indicating from which population the observation arose. Theresulting likelihood function is interpreted in terms of the conditional probability density of a sample from a mixture of populations, given the identification parameter of each observation. Clustering algorithms are obtained by applying a method of iterated maximum likelihood to this like-lihood function.

Analyzing Free-Response Data in Marketing Research:

Abstract: This article presents a new approach to the analysis of free-response data matrices using reduced-space and clustering procedures. Two marketing applications are presented; limitations and possible...

Passage organization and imposed learning strategies in comprehension and recall of connected discourse.

Abstract: Ninety-nine Ss were assigned randomly to learn a passage comprised of six paragraphs in which the statements were organized by concept name, or concept attribute, or in which the statements were scrambled. Each complete passage contained the same 36 statements. These treatments were orthogonally crossed with instructions to employ a name clustering strategy, an attribute clustering strategy, or a subjectively determined organizing strategy. Three learning trials were administered, each of which was followed by free recall. Passages organized by concept name were found to result in greater recall than passages organized by attributes. The name clustering strategy was more dominant than the attribute clustering strategy, lncongruence between passage organization and advocated clustering strategy resulted in greater recall than did congruency. Implications of these results for cognitive processing of information are discussed

An Algorithm for Nonsupervised Pattern Classification

Abstract: An algorithm for classifying a data set into an initially unknown number of categories is presented. It is composed of procedure for selecting initial points, a mode estimation procedure, and a classification rule. An integer valued function is defined on the sample space and a gradient search technique is used for estimating its modes. A procedure for mode estimation in the case of an infinite data set is also proposed. Sufficient conditions for the convergence to the neighborhood of the modes have been stated. The algorithm was used for clustering multicategory artificially generated data sets and was compared with an optimal classification scheme.

A districting experiment with a clustering algorithm

Abstract: In this paper, we consider the problem of how to partition a geographic area into a collection of contiguous, approximately equal population districts. This problem is analogous to the graph theory problem of partitioning a network into a fixed number of contiguous sections such that the sum of the node weights within each sector is the same. In experimenting with this heuristic algorithm, we found that the procedure is extremely fast computationally, does a reasonable job of partitioning the network, but does not terminate at what could be called a strong set of local optima. (The question of when local optima become global optima, as well as the various properties of heuristic algorithms that this question implies, is being considered by s. Savage in his Ph.D. thesis at Yale University.) Thus, the algorithm terminates at one out of many solutions that do approximately the same job. The heuristics that make up this algorithm, however, are not discriminating enough to pick out the best of these solutions. Nevertheless, we derive some very useful benefits from this procedure. First, the procedure quickly makes a good initial guess regarding an optimal partition. This guess can then be used as an upper bound to make exact, but more time-consuming, algorithms more efficient. Second, since the algorithm generates many good, approximately equal partitions cheaply in terms of computer time, each of these solutions can then be examined for other attributes (such as compactness) and the best of these solutions accepted. Other attempts to do redistricting and graph partitioning have been made in the past. There is also a close relationship between the redistricting problem and a class of facility location problems. Some of the studies in these areas are Hess and Weaver,' Garfinkel,2 M a r a ~ a n a , ~ Trehan,4 Teitz and Bart,5 Marks and colleagues,6 Spielberg?, * Scott,g and Revelle and Swain.Io

A clique-detection algorithm based on neighborhoods in graphs

Abstract: The notion of cliques is the logical starting point for a broad class of clustering problems. Algorithms for the identification of the cliques of a graph are of great importance in automatic classification by computer. This paper presents an approach which proceeds recursively from the given graph downward through a chain of subgraphs. The proposed approach permits the removal of more than a single point in the progression from one subgraph to the next, thus reducing the number of iterations required for completion.

The role of taximetrics in angiosperm taxonomy i. empirical comparisons of methods using crotalaria l.

Abstract: SUMMARY Two hundred and seventy-three African species of Crotalaria L. (Leguminosae) were subjected to three taximetric procedures: single-link clustering, with subgraph representation; principal co-ordinate analysis; and median clustering. The classifications suggested by these methods were compared with each other, compared with a contemporary orthodox classification by Polhill and checked against the original plant material to see if the groupings were recognizable notwithstanding any similarity with Polhill's classification. Above the species level Crotalaria presents classification problems in size and complexity not uncommon for the angiopersms. The results of single-link clustering suggest a classification close to Polhill's. In one case, where selected characters had been used, much of the correspondence is exact and of the few discrepancies some reveal possible improvements to the classification and others were caused by external factors. The principal co-ordinate analysis results are difficult to interpret, suggest some unrecognizable groupings and only vaguely resemble Polhill's arrangement. The median clustering results suggest a tidy-looking classification which, on investigation, proves largely spurious. The single-link and principal co-ordinate analysis were used to process two different sets of data. One set consisted of data for characters chosen mainly from the other set, selected for high intercorrelations using the information contribution measure. For both methods data for the selected character set gave more useful results. These findings favour the use of single-link clustering for classifying angiosperms above the species level. They are a variance with the conclusions of some earlier empirical comparisons using angiosperms, but lend support to a number of theoretical arguments for preferring the single-link procedure.

Function-Point Cluster Analysis

Abstract: Katz, Jeffrey and F. James Rohlf (Department of Ecology and Evolution, State University of New York, Stony Brook, New York 11790) 1973. Function-point cluster analysis. Syst. Zool., 22:295-301.-The gradient clustering method of Ihm (1965) was reinvestigated and applied to several sets of real and artificial data. It is based on the technique of defining a function which has the property of being maximal in regions where there are high densities of points and low elsewhere. Points are considered to be in the same cluster if they are "under" the same local maximum of this function. The clusters obtained at different hierarchic levels are not necessarily nested. A generalization of the skyline graph is presented to depict such a system of clusters. [Clustering.]

Nonhierarchical clustering methods

Abstract: The non-hierarchical clustering methods are designed to cluster data units into a single classification of k clusters, where k either is specified a priori or is determined as part of the clustering method. The idea in most of these methods is to choose some initial partition of the data units and then alter cluster memberships to obtain a better partition. The various algorithms that have been proposed differ as to what constitutes a “better partition” and the methods that might be used for achieving improvements. The non-hierarchical clustering methods might be used with larger problems than the hierarchical methods because it is not necessary to calculate and store the similarity matrix; it is not necessary to store the data set. The data units are processed serially and can be read from tape or disk, as needed. This characteristic makes it possible to cluster arbitrarily large collections of data units.

The JSC Clustering Program ISOCLS and its Applications

Abstract: The clustering program ISOCLS developed at the Johnson Space Center, Houston, Texas, has been extensively used in the pattern analysis and classification of remote sensor data collected by aircraft and by the Earth Resources Technology Satellite ERTS-1. This paper discusses the theory behind this clustering algorithm. Several new ideas that have been incorporated in ISOCLS are discussed. Among these are the novel philosophy of operation behind the procedure, which assumes that a population (i.e., a class or a cluster) can be treated as the union of an appropriate number of subpopulations, and the termination of the clustering program by a 'chaining algorithm.' Finally, this paper reports the results of the application of ISOCLS to an investigation on rangeland vegetation mapping using ERTS-1 data.

Development of New Pattern-Recognition Methods.

Abstract: : The problem in pattern recognition is to find a classification or description of the data patterns that matches or suits the data. New methods in pattern recognition are studied in relation to classical approaches using techniques of multivariate statistical analysis. The application of these techniques to specific problems in physical, engineering, behavioral, and other sciences is reviewed. The problems of improved data description and dimensionality reduction are tackled by means of clustering approaches. Several improved clustering methods are developed for general pattern recognition: a new approximate procedure for computing the minimal-spanning tree, a new application of the Kolmogorov-Smirnov test for cluster validity, and a new application of relativistic principles in measures of relationship. Experiments using interactive graphic displays to illustrate these new methods are described, and application of computer programs to meteorological problems is demonstrated. (Modified author abstract)

The Dynamic Clusters Method and Optimization in Non-Hierarchical Clustering

Abstract: Algorithms which are operationnally efficient and which give a good partition of a finite set, produce solutions that are not necessarily optimum. The main aim of this paper is a synthetical study of properties of optimality in spaces formed by partitions of a finite set. We formalize and take for a model of that study a family of particularily efficient techniques of "clusters centers" type. The proposed algorithm operates on groups of points or "kernels"; these kernels adapt and evolve into interesting clusters.

Dynamic Scheduling of Large Digital Computer Systems Using Adaptive Control and Clustering Techniques

Abstract: This research is directed toward the development of a scheduling algorithm for large digital computer systems. To meet this goal methods of adaptive control and pattern recognition are applied. As jobs are received by the computer, a pattern recognition scheme is applied to the job in an attempt to classify its characteristics, such as a CPU-bound job, an I/O job, a large memory job, etc. Simultaneously, another subsystem, using a linear programming model, evaluates the overall system performance, and from this information an optimized (or desired) job stream is determined. When the processor requests a new job, it is chosen from the various classifications in an attempt to meet the optimal (or desired) job stream. After the jobs are completely processed, their characteristics are compared to the projected classification produced by the pattern discriminant function. The results are then returned to the discriminant function to update the decision mechanism, a minimum-distance discriminant function. From a systems point of view, this results in an adaptive or self-organizing control system. The overall effect is a dynamic scheduling algorithm. Simulation studies indicated that the scheduler was able to adapt to changing work loads, and it improved the turnaround times significantly. These simulation studies were based on a multiprocessor-uniprogram environment.

An Unsupervised Classification Technique for Multispectral Remote Sensing Data

Abstract: Description of a two-part clustering technique consisting of (a) a sequential statistical clustering, which is essentially a sequential variance analysis, and (b) a generalized K-means clustering. In this composite clustering technique, the output of (a) is a set of initial clusters which are input to (b) for further improvement by an iterative scheme. This unsupervised composite technique was employed for automatic classification of two sets of remote multispectral earth resource observations. The classification accuracy by the unsupervised technique is found to be comparable to that by traditional supervised maximum-likelihood classification techniques.

Classification of Scientific Documents by Means of Self-Generated Groups Employing Free Language.

Abstract: A study was undertaken to classify mechanically a document collection using the free-language words in the titles and abstracts of a corpus of 261 physics research papers. Using a clustering algorithm, results were obtained which closely duplicated the clusters obtained by previous experiments with citations. A brief comparison is made with a traditional manual classification system. It is shown that the mechanical procedure is capable of achieving simultaneous average relevance and recall figures above 80%.

Unsupervised spatial clustering with spectral discrimination

Abstract: The development of a computer program is reported for extracting features from remotely sensed data presented in digital image form. This computer program requires no human supervision or prejudgment and operates unassisted on the raw digital data. A condensed general background is included on remote sensing of earth features and a short synopsis on some of the most commonly used types of feature extraction techniques. Results obtained from the unsupervised feature extraction computer program along with a description and listing of the computer program are presented.

Clustering of one-dimensional ordered data

Abstract: A set ofn ordered real numbers is partitioned by complete enumeration intok clusters such that the sum of the sum of squared deviations from the mean-value within each cluster is minimized.