A Network Intrusion Detection Scheme by using Improved Loss Compensated Deep AutoEncoder

In the current digital era, with the rapid popular-ization of network technology, it brings plenty of conveniences to the public. However, the network is attacked more and more frequently, and the network security problem has become a primary concern of the users. To improve network security, as one of the most important means of active defense in network security field, network intrusion detection has been widely studied, which aims to identify and prevent potential security attacks through monitoring and analyzing the network traffic. However, existing network intrusion detection technologies usually adopts a two-stage detection mode of feature extraction followed by detection. This split detection mode makes it impossible for the detector to alleviate the feature loss arises from feature extraction by the way of optimization and other methods, thereby lowering the detection efficiency and accuracy. For solving the above problem, we present a novel network intrusion detection method. We firstly design a novel deep learning model called Improved Loss Compensated Deep AutoEncoder (ILC-DAE) by improving the traditional Loss Compensated Deep AutoEncoder (LC-DAE). ILC-DAE can perform feature extraction and traffic detecting at the same time. Then, we applied ILC-DAE to design a new method for network intrusion detection, which can enhance the intrusion detection accuracy and efficiency. At last, we implement our ILC-DAE network intrusion detection method and deploy a prototype system, in which the effectiveness of our ILC-DAE network intrusion detection method will be verified. Experimental results and analysis prove that compared with the existing intrusion detection methods, our ILC-DAE network intrusion detection method obtains better intrusion detection performance.