Anomaly Detection Using Multi-Level and Multi-Dimensional Analyzing of Network Traffic

With the rapid growth of the categories and numbers of network attacks and the increasing network bandwidth, network traffic anomaly detection systems confront with both higher false positive rate and false negative rate. A traffic anomaly detection system with high precision is presented in this paper. Firstly, we use multi-level and multi-dimensional online OLAP method to analyse traffic data. In order to reduce the computational and space complexity in this analytical process, some optimization strategies are applied in building DetectCube, the minimal directed Steiner tree algorithm is adapted to optimize multiple query on the Cube, and the traffic data is summarized at appropriate level with the help of discovery-driven exploration method. Secondly, a concept of entropy to measure the distribution of traffic on some particular dimensions is given and the values of entropy in every window and every Group-By operation are collected to form multiple time series of entropy. Finally, we employ one-class support vector machine to classify this multi-dimensional time series of entropy to achieve the purpose of anomaly detection. The proposed traffic anomaly detection system is validated and evaluated by comparing it with existed systems derived from a lot of real network traffic data sets. Our system can detect attacks with high accuracy and efficiency.