Improving Supervised Outlier Detection by Unsupervised Representation Learning and Generative Adversarial Networks

In this paper, a semi-supervised outlier detection algorithm is presented. The proposed method is an extension of XGBOD (Extreme Gradient Boosting Outlier Detection) developed by Zhao and Hryniewicki and combines the benefits of both unsupervised representation learning and generative adversarial networks. The previous studies show that multiple unsupervised outlier detection algorithms extract useful representations from the underlying data. The results presented herein show that the augmentation of their extracted representations by GANs (Generative Adversarial Networks) further improves the performance of supervised outlier detection on several datasets.