Anomaly Detection Using Replicator Neural Networks Trained on Examples of One Class

Abstract. Anomaly detection aims to find patterns in data that are significantly different from what is defined as normal. One of the chal-lenges of anomaly detection is the lack of labelled examples, especially for the anomalous classes. We describe a neural network based approach to detect anomalous instances using only examples of the normal class in training. In this work we train the net to build a model of the normal examples, which is then used to predict the class of previously unseen instances based on reconstruction error rate. The input to this network is also the desired output. We have tested the method on six benchmark data sets commonly used in the anomaly detection community. The re-sults demonstrate that the proposed method is promising for anomaly detection. We achieve F-score of more than 90 % on 3 data sets and out-perform the original work of Hawkins et al. on the Wisconsin breast cancer set