A Novel Preprocessing Methodology for DNN-Based Intrusion Detection

Intrusion Detection System (IDS) plays a vital role in cybersecurity. One major challenge in state-of-the-art IDS is that the Advanced Persistent Threats (APTs) cannot be well detected. Deep Neural Networks (DNNs) extract latent features of input data, which is considered to be able to capture intrinsic property of APTs, to detect APTs. However, DNN is not well used in APT detection. In this paper, a DNN model is utilized for intrusion detection and a novel method of preprocessing is proposed to improve the performance of the detection algorithms. We consider the statistical characteristics of training set to be the statistical characteristics of test set and detection set. Experiment results show that our preprocessing methodology achieves a better performance in terms of accuracy, recall and F1 Score than traditional preprocessing methods, which means a better detection for APTs.