Showing papers on "Feature selection published in 1987"

Feature Selection for Automatic Classification of Non-Gaussian Data

Abstract: A computer-based technique for automatic selection of features for the classification of non-Gaussian data is presented The selection technique exploits interactive cluster finding and a modified branch and bound optimization of piecewise linear classifiers The technique first finds an efficient set of pairs of oppositely classified clusters to represent the data Then a zero-one implicit enumeration implements a branch and bound search for a good subset of features A test of the feature selection technique on multidimensional synthetic and real data yielded close-to-optimum, and in many cases optimum, subsets of features The real data consisted of a) 1284 12-dimensional feature vectors representing normal and abnormal breast tissue, extracted from X-ray mammograms, and b) 1060 30-dimensional feature vectors representing tanks and clutter in infrared video images

A fast model selection procedure for large families of models

Abstract: An efficient procedure for model selection from large families of models is described. It is closely related to the all possible models approach but is considerably faster. It is based on two principles: first, if a model is accepted, then all models that include it are considered to be accepted; second, if a model is rejected, then all of its submodels are considered to be rejected. Application of the procedure to variable selection in multiple regression is illustrated. General algorithms are described that enable the procedure to be applied to any family of models that forms a lattice. As an example, a problem in multiple comparisons is considered.

Feature selection for texture recognition based on image synthesis

Abstract: An efficient method for selection of features suitable for classification of textured images is presented. The spatial interaction of gray levels in a local neighbourhood N is modeled by stochastic random field models. The estimates of the model parameters are taken as textural features denoted by fN. Selection of an N that would yield powerful features is done through visual examination of images synthesized using fN. Experimental studies involving nine different types of natural textures yield 97% classification accuracy.

Advances in statistical pattern recognition

Abstract: Statistical pattern recognition is now a mature discipline which has been successfully applied in several application domains. The primary goal in statistical pattern recognition is classification, where a pattern vector is assigned to one of a finite number of classes and each class is characterized by a probability density function on the measured features. A pattern vector is viewed as a point in the multidimensional space defined by the features. Design of a recognition system based on this paradigm requires careful attention to the following issues: type of classifier (single-stage vs. hierarchical), feature selection, estimation of classification error, parametric vs. nonparametric decision rules, and utilizing contextual information. Current research emphasis in pattern recognition is on designing efficient algorithms, studying small sample properties of various estimators and decision rules, implementing the algorithms on novel computer architecture, and incorporating context and domain-specific knowledge in decision making.

Variable selection in clustering and other contexts

Abstract: An edge connector assembly for effecting electrical connection to a plurality of connection tabs arranged along and adjacent an edge of an inserted printed circuit board. The connector assembly generally includes an insulated housing and a plurality of like spring terminals mounted in the housing. Each terminal has a contact portion engaging a connection tab. The housing includes a first wall having a front portion overlying the board and a rear portion. A second wall is parallel to and spaced from the first wall and is joined to the rear portion thereof. A lip support extends from the second wall canted away from the first wall. An edge receiving slot is defined between the lip support and the overlying portion of the first wall for receiving the edge of the board therein. A plurality of terminal receiving cavities are formed between the first and second walls, each cavity including terminal mounting means on the first wall for mounting the terminals so that the contact portion thereof is disposed in the edge receiving slot for engagement with the connection tabs. The board is rotatable within the edge receiving slot from a non-engaging position wherein the contact portions are spaced from the connection tabs to a mounted position where the contact portions are engaging the connection tabs. Locking means are formed on the overlying portion of the first wall for cooperating with the board for holding the board in the mounted position.

The use of conditional cutoffs in a forward selection procedure

Abstract: Using a forward selection procedure for selecting the best subset of regression variables involves the calculation of critical values (cutoffs) for an F-ratio at each step of a multistep search process. On dropping the restrictive (unrealistic) assumptions used in previous works, the null distribution of the F-ratio depends on unknown regression parameters for the variables already included in the subset. For the case of known σ, by conditioning the F-ratio on the set of regressors included so far and also on the observed (estimated) values of their regression coefficients, we obtain a forward selection procedure whose stepwise type I error does not depend on the unknown (nuisance) parameters. A numerical example with an orthogonal design matrix illustrates the difference between conditional cutoffs, cutoffs for the centralF-distribution, and cutoffs suggested by Pope and Webster.

Feature selection in colour images for token matching

Abstract: We are currently developing an algorithm for knowledge-based corre­ spondence analysis in dynamic stereo images. In this paper we describe the feature selection algorithms and the relational data structure used for matching. We also present a clustering algorithm that is used to group the tokens into sets of similar tokens. The matching algorithm uses this classification information to plan a matching strategy. The least ambiguous image features are matched first and are used as a 'han­ dle' for constraint propagation. We present two algorithms for feature selection in colour images —a region grower and an interest opera­ tor, as well as experimental results for point classification and region classification.

Variable Selection in Logistic Regression.

Abstract: : In many situations, we are interested in selection of important variables which are adequate for prediction under a logistic regression model In this paper, some selection procedures based on the information theoretic criteria are proposed, and these procedures are proved to be strongly consistent Keywords: Maximum likelihood estimation; Multivariate analysis; Asymptotic expansion

Analysis of the Performance of a Parametric and Nonparametric Classification System: An Application to Feature Selection and Extraction in Radar Target Identification.

Abstract: : This dissertation investigates new parametric and nonparametric bounds on the Bayes risk that can be used as a criterion in feature selection and extraction in radar target identification (RTI). For the parametric case, where the form of the underlying statistical distributions is known, Bayesian decision theory offers a well-motivated methodology for the design of parametric classifiers. This investigation provides new bounds on the Bayes risk for both simple and composite classes. Bounds on the Bayes risk for M classes are derived in terms of the risk functions for (M-1) classes, and so on until the result depends only on the Pairwise Bayes risks. When the parameters of the underlying distributions are unknown, an analysis of the effect of finite sample size and dimensionality on these bounds is given for the case of supervised learning. For the case of unsupervised learning, the parameters of these distributions are evaluated by using the maximum likelihood technique by means of an iterative method and an appropriate algorithm. Finally, for the nonparametric case, where the form of the underlying statistical distributions is unknown, a nonparametric technique, the nearest-neighbor (N N) rule, is used to provide estimated bounds on the Bayes risk. Two methods are proposed to produce a finite size risk close to the asymptotic one. The difference between the finite sample size risk and the asymptotic risk is used as the criterion of improvement.

Feature Selection Applied to Radar Target Identification.

Abstract: : Three feature selection algorithms are investigated and applied to characterize optimum sets of frequencies for radar target identification. One algorithm is of the nonparametric discriminant analysis type, the other two algorithms, the pairwise exponential weight distance algorithm and the pairwise probability of error algorithm, are parametric and incorporate information about the measurement noise into the feature selection process. The utility of these feature selection algorithms for radar target identification is then evaluated through Monte-Carlo simulations. It is found that significant gain in classification performance can be achieved by using the optimum sets of frequencies characterized by the parametric algorithms.

A practical variable for generalized selection method least squares and its application

Abstract: The j·th best subset problem for the generalized least squares is formulated in which statistical criteria as well as non-statistical conditions are introduced. Non-statistical conditions are based on a knowledge of the scientific field to which research is related, natural logic and common sense, while statistical criteria are t-test, Durbin-Watson serial correlation test, absolute relative error test, turning point test and fitting test, depending on the covariance matrix of a disturbance term and type of data. Various technical methods are devised to make a computer solve the first (j=l) to the J-th (e.g., J=10) best subset problems in one computer-run, depending on whether or not a researcher has a new criterion or appropriate values for the parameters used to evaluate meaningful subsets before estimation. Then, the ultimately best subset among the best J subsets is regarded as a practical solution to the variable selection problem for the generalized least squares. The System OEPP can handle the proposed variable selection method.