Showing papers on "Feature selection published in 1982"

Variable selection techniques in discriminant analysis: II. Allocation

Abstract: The literature dealing with the selection of variables for allocation procedures is examined. Two types of technique are discussed, namely, those which use group separation as the criterion for variable selection and those which more appropriately employ error rates in allocation.

Variable selection techniques in discriminant analysis: I. Description

Abstract: The problem of assessing the relative importance of variable subsets in discriminant analysis is discussed. Attention is focused on techniques for determining those subsets which are ‘adequate’ for discrimination from the descriptive viewpoint. A number of procedures which have been proposed or used in the literature are described, illustrated and compared with reference to various aspects, including rationale, statistical significance testing and computational difficulties.

Curve fitting and modeling with splines using statistical variable selection techniques

Abstract: The successful application of statistical variable selection techniques to fit splines is demonstrated. Major emphasis is given to knot selection, but order determination is also discussed. Two FORTRAN backward elimination programs, using the B-spline basis, were developed. The program for knot elimination is compared in detail with two other spline-fitting methods and several statistical software packages. An example is also given for the two-variable case using a tensor product basis, with a theoretical discussion of the difficulties of their use.

36 Topics in measurement selection

Abstract: Publisher Summary One of the major problems one encounters during the design phase of an automatic pattern recognition system is the identification of a good set of measurements. These measurements, to be performed on future unclassified patterns, should enable the recognition system to classify the patterns as correctly as possible. The identification of a measurement set is generally carried out in two phases. The first phase, pattern analysis, uses a variety of techniques that allow the designer to explore raw pattern data, and to infer some of its structure. In the second phase, commonly called feature selection or measurement selection, the preliminary measurement set must be reduced in size to meet the cost/performance trade-off. Various techniques exist to achieve the data reduction called for. One can characterize these techniques either according to the way in which the data reduction was achieved or according to the purpose of the reduced data. This chapter considers the measurement selection techniques of the discriminating category that use the selection method as a means for data reduction.

Speaker recognition using a feature weighting technique

Abstract: This paper describes a technique for increasing the ability of a text-dependent speaker recognition system to discriminate between speaker classes; this technique is to be performed in conjunction with the nonlinear time alignment between a reference pattern and a test pattern. Unlike the standard approach, where the training of the recognition system merely consists of storing and averaging or selecting the time normalized training patterns separately for each class, the training phase of the system is extended in that a weight is determined for each individual feature component of the complete reference pattern according to the ability of the feature to distinguish between speaker classes. The weights depend on the time axis as well as on the frequency axis. The overall distance computed after nonlinear time alignment between a reference pattern and a test pattern thus becomes a function of the given set of weights of the reference class considered. For each class, the optimum weights result from the ideal criterion of minimum error rate. Instead of this criterion, the closely related but mathematically more convenient Fisher criterion is used that leads to a closed from solution for the unknown weights. Based on these weights, the selection of subsets of effective features is studied in order to further improve the class discrimination. The feature weighting and selecting techniques are tested using a data base of utterances recorded off dialed-up telephone lines. The experiments indicate that feature weighting and feature selection can reduce the error rates by a factor of two or more both for speaker identification and speaker verification.

Signal classification for automatic industrial inspection

Abstract: Details of a system designed to inspect web-type products are presented. The sensing device used is a laser scanner, and a brief description of the scanner is given. A set of bandpass filters is proposed as feature extracters. The required characteristics of the filters are determined through digital simulation, using feature selection methods. The linear classifier is designed from a set of training signals. Contextual information is used in the classification of signals into good product or into various defective categories. In addition, results of a study on inspection of magnetic tapes and abrasive sheets are described.

Linear And Nonlinear Feature Extraction Methods Applied To Automatic Classification Of Proliferating Cells

Abstract: Numerical experiments were performed to find optimum feature extraction procedures for the classification of mouse L-fibroblasts into Gl, S and G2 subpopulations. From images of these cells different feature sets such as geometric, densitometric, textural and chromatin features were derived which served as data base for the numerical experiments. Linear and nonlinear supervised stepwise learning techniques for the discrimination of the cells into Gl, S and G2 were performed. The classification error was used as criterion for the evaluation of the different numerical feature selection methods. Optimum results were obtained by combining distance based feature selection methods with nonlinear discriminant analysis. The successive solution of 2-class problems improves the results compared to the solution of the 3-class problem. Linear discriminant analysis then may surpass quadratic discriminant analysis.© (1982) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

A Forward Multivariate Regression Procedure Based on the Maximization of the RV Coefficient

Abstract: The RV Coefficient is shown to be an acceptable criterion for variable selection in multivariate linear regression. An efficient computational procedure for forward selection based on this criterion is detailed. Finally, an example is given.

Distance Measures of Distributions and Classification Oriented Feature Selection

Abstract: Distance measures of distributions are often used to estimate upper and lower bounds on the probabilities of misclassification. Sharp lower and upper bounds are of great importance for feature selection, that means for classification oriented feature interpretation. MATUSITA affinity/6/ gives sharp upper bounds, the divergence /4/ lower bounds on the probabilities of misclassification. This paper discusses the properties of these two distance measures. Other measure are compared at length in /9/.© (1982) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.