Uncertainty assessment using neural networks and interval neutrosophic sets for multiclass classification problems

Kraipeerapun, P., Fung, C.C. and Wong, K.W. (2007) Uncertainty assessment using neural networks and interval neutrosophic sets for multiclass classification problems. WSEAS Transactions on Computers, 6 (3). pp. 463-470.

Preview

PDF - Published Version Download (1082kB) | Preview

Abstract

This paper presents an approach to multiclass classification. A pair of k-class neural networks are trained to predict k pairs of truth membership and false membership values. The k pairs of errors in the prediction of unknown input patterns are also estimated using interpolation techniques. Two techniques are proposed for the multiclass classification in this paper. First, the truth and false memberships are compared in order to classify the input pattern into multiple classes. Second, estimated errors are used to weight the degrees of truth and false memberships. After that, the results of the combination between weighted truth and weighted false memberships are used for the classification. The estimated errors as well as the difference between the truth and false membership values are considered as the elements of the indeterminacy membership used to identify level of uncertainty in the multiclass classification. Together the three membership values form interval neutrosophic sets. We experiment our technique to the classical benchmark problems including balance, ecoli, glass, lenses, wine, yeast, and zoo from the UCI machine learning repository. Our approach improves classification performance compared to an existing technique which applied only to the truth membership created from a single k-class neural network.