Classifying patterns into two classes is a typical problem of binary classification in pattern recognition. Binary classification is an industrial problem in many fields like medicine, search mechanism, diagnostic of disease in humans, security and many other aspects. In this paper, we have proposed a random subspace based ensemble data dependent classification model for the binary classification problem. The proposed method makes use of the information about the structure of given data and the availability of the training instances, in selection of the classification model. A subspace ensemble for a set of one class and two-class classifier are generated, trained and tested on the given data. The proposed method is evaluated on receiver operating curve (ROC), cross validation accuracy and Q-statistics. From the empirical results obtained, we have reached the following conclusions that the overall performance of the two class ensemble was better because of the ability of the ensemble to make use of the knowledge of both the positive and negative samples and thus constructs better class boundaries. The one class ensemble makes use of positive samples only and gives better performance when (i) training patterns are sparse and (ii) outlier detection is required.
Key words: One class classifier, two-class classifier, binary classification, classification model, receiver operating curve (ROC), evaluation, Q- statistics.
Copyright © 2023 Author(s) retain the copyright of this article.
This article is published under the terms of the Creative Commons Attribution License 4.0