Single-Class Classifier Learning using Neural Networks: Extracting Context from Unlabeled Data

A. Skabar (Germany)


Neural networks, classification


Single-class classifier learning is the problem of learning a classifier from a set of training examples in which only examples of the target class are present. Most existing approaches to this problem are based on density estimation and hence suffer from the usual problems associated with estimating probability densities in high dimensional spaces. This paper describes how feedforward neural networks can be used to learn a classifier from a dataset consisting of (labeled) examples of the target class (positive examples) together with a corpus of unlabeled (positive and negative) examples. Results obtained from applying the technique to several datasets from the UCI repository demonstrate classification performance comparable to that achieved using conventional supervised learning. The technique is applicable to a broad range of classification and pattern recognition problems in which either the nature of the problem do main, or the expense of labeling the data, make it difficult to supply labeled counter-examples.

Important Links:

Go Back