Directly Identify Unexpected Instances in the Test Set by Entropy Maximization

In real applications, a few unexpected examples unavoidably exist in the process of classification, not belonging to any known class. How to classify these unexpected ones is attracting more and more attention. However, traditional classification techniques can’t classify correctly unexpected instances, because the trained classifier has no knowledge about these. In this paper, we propose a novel entropy-based method to the problem. Finally, the experiments show that the proposed method outperforms previous work in the literature.