Outliers' effect reduction of one-class neural networks classifier

The One Class Auto Associative Neural Network (AANN) has been investigated for solving various problems. Nonetheless, it is sensitive to the presence of outliers in the training set, which is known problem for one-class classifiers. For this, attempts have been done via proposing the use of efficient kernel and ensemble method to reduce the effect of outliers for one class support vector machine classifier. However, for the AANN, even with ensemble method, the effect of outliers is still maintained. Thus, we propose in this paper the joint use of ensemble method with a selection algorithm to select the appropriate training samples for the AANN, which leads to better reduction of the outliers' effect and therefore improving the AANN ensemble and classification robustness. Experimental results conducted on several real-world datasets prove the effective use of the proposed approach.