A Drift-Aware Online Learner for Anomaly Detection from Streaming Data

Streaming data has been evolved in a dynamically changing and evolving environment. Therefore, concept drift or changing the underlying distribution of data over time is considered as an important challenge in processing this type of data. Moreover, concept drift affects the performance of anomaly detection process. The problem of anomaly detection in streaming data is applied to many important applications, for instance, intrusion detection in computer networks or traffic management in the road networks. In recent years, some tensor decomposition based approaches have been presented that track the main pattern or subspace of data in an online manner and adapt the learner with probabilistic changes continuously in all time-intervals by using an implicit strategy. We propose an online approach that detects the concept drift in an explicit manner. Moreover, the learner has been adapted with drift and changes only in their occurrences using informed strategy. Evaluation of the proposed method is performed with real datasets. Analysis of the obtained results confirms the promising performance of the proposed method in terms of learning and detection.