DeepWindow: An Efficient Method for Online Network Traffic Anomaly Detection

Zhenping Shi, Jie Li, Chentao Wu, Jinyuan Li

Abstract: With the explosion of network traffic volume, high efficient and large-scale network traffic anomaly detection methods becomes necessary. However, existing methods often fail to take into account both the detection delay and the detection accuracy. We propose a novel method, focusing on period-wise detection. We use Long Short-Term Memory (LSTM) to establish abnormal traffic detection model. Besides, We use some big data processing frameworks for online network traffic collection and preprocessing. Performance evaluation shows that our online anomaly detection model outperforms other anomaly detection methods based on traditional anomaly detection methodologies.