Showing papers on "Feature selection published in 1974"

Pattern recognition principles

Abstract: The present work gives an account of basic principles and available techniques for the analysis and design of pattern processing and recognition systems. Areas covered include decision functions, pattern classification by distance functions, pattern classification by likelihood functions, the perceptron and the potential function approaches to trainable pattern classifiers, statistical approach to trainable classifiers, pattern preprocessing and feature selection, and syntactic pattern recognition.

Nonlinear Intrinsic Dimensionality Computations

Abstract: In pattern recognition, the raw data and dimensionality of the measurement space is usually very large. Therefore, some form of dimensionality reduction has been commonly considered as a practical preprocessing method for feature selection. Based on a method that increases the variance while maintaining local structure, a technique is developed to determine intrinsic dimensionality. A cost function is introduced to guide the maintenance of the rank order and therefore local structure. Two criteria of using the cost function to increase the variance have been introduced. Several methods of defining the local regions are suggested. A program is implemented and tested to find the intrinsic dimensionality of a variety of experimental data.

Computerized pattern recognition applications to chemical analysis. Development of interactive feature selection methods for the K-nearest neighbor technique

Abstract: : A systematic approach has been developed for feature selection in the application of the K-nearest neighbor (KNN) computerized pattern recognition method. The approach uses an operator-interactive computer system. A large number of potentially-useful features for classification of patterns can be screened for the most relevant members by a combination of recommended procedures. These include: (a) one-dimensional KNN classification of all patterns using each feature individually; (b) inspection of histogram displays of classification records for each feature; and (c) establishment of consensus classifications from combined one-dimensional results. A computerized trial-and-error procedure can then be implemented to find the best combination of a minimum number of features for accurate classification using the multi-dimensional KNN method. (Modified author abstract)

Cluster Analysis Based on Dimensional Information with Applications to Feature Selection and Classification

Abstract: A new clustering algorithm is presented that is based on dimensional information. The algorithm includes an inherent feature selection criterion, which is discussed. Further, a heuristic method for choosing the proper number of intervals for a frequency distribution histogram, a feature necessary for the algorithm, is presented. The algorithm, although usable as a stand-alone clustering technique, is then utilized as a global approximator. Local clustering techniques and configuration of a global-local scheme are discussed, and finally the complete global-local and feature selector configuration is shown in application to a real-time adaptive classification scheme for the analysis of remote sensed multispectral scanner data.

Variable Selection Technique for Classification Problems.

Abstract: This paper illustrates the usefulness of a variable selection technique when the linear discriminant function is used to develop a classification rule. A stepwise technique is given together with an illustration of how the expected loss can be reduced using fewer variables.

Suboptimum Linear Feature Selection in Multiclass Problem

Abstract: A suboptimum method of linear feature selection in multiclass problem is presented. The set of features is selected in sequential manner based on an upper bound on the probability of error. The proposed method is applied to a problem of classifying Japanese vowels. Computer simulation results are presented and discussed.

Feature selection and the concept of immediate neighborhood in the context of clustering techniques

Abstract: The concept of feature selection in a nonparametric unsupervised learning environment is practically undeveloped because no true measure for the effectiveness of a feature exists in such an environment. The lack of a feature selection phase preceding the clustering process seriously affects the reliability of such learning. New concepts such as significant features, level of significance of features, and immediate neighborhood are introduced which result in meeting implicitly the need for feature slection in the context of clustering techniques.

A Minimum-Cost Feature-Selection Algorithm for Binary-Valued Features

Abstract: An algorithm to select the minimum-cost collection of binary-valued features for use with a linear pattern classifier is presented. The feature-selection algorithm is motivated by the convex-hull representation of pattern-space separability. Combinatorial analysis and linear programming are used to find the minimum-cost collection of binary-valued features associated with a given set of preclassified patterns. A description of the interaction between these algorithm components is provided. The algorithm guarantees that its optimal feature set will correctly classify every pattern in the classifier's training sample. Coinputational considerations associated with algorithm use are discussed. An application of the algorithm to a three-feature classifier is presented in detail.

A Binary Feature Extraction Technique

Abstract: Numerous schemes are available for feature selection in a pattern recognition problem, but the feature extraction process is largely intuitive. A sequential feature extraction scheme is proposed for binary features. A decision function, which is linear and near optimal, is developed concurrently with each feature. Performance bounds are developed for several design strategies. Experimental results are given to illustrate the use of the scheme and the effectiveness of the performance bounds.

Some results on the divergence criterion used to solve the feature selection problem

Abstract: Techniques for solving the feature selection problem are presented. Topics discussed include the reduction of the number of variables in "best b", and the iterative selection of H sub i.

On differentiating the probability of error in the multipopulation feature selection problem, 2

Abstract: A condition for the Gateaux differentiability of the probability of misclassification as a function of a feature selection matrix B, assuming a maximum likelihood classifier and normally distributed populations, is given. It is also shown that if the probability of error has a local minimum at B then it is differentiable at B.

Results on the two population feature selection problem using probability of correct classification as a criterion

Abstract: Variational equations are presented for maximizing the probability of correct classification as a function of a 1xn feature selection matrix B for the two-population problem. For the special case of equal covariance matrices the optimal B is unique up to scalar multiples and rank one sufficient. For equal population means, the best 1xn B is an eigenvector corresponding either to the largest or smallest eigenvalue of sigma sub 2 to the minus 1 power sigma sub 1 where sigma sub 1 and sigma sub 2 are the nxn covariance matrices of the two populations. The transformed probability of correct classification depends only on the eigenvalue. Finally, a procedure is proposed for constructing an optimal or nearly optimal kxn matrix of rank k without solving the k-dimensional variational equation.