Feature-oriented Design of Visual Analytics System for Interpretable Deep Learning based Intrusion Detection

Deep Learning models have demonstrated significant performance on different tasks such as computer vision, natural language processing, etc. In recent years, these models have also achieved remarkable progress in Intrusion Detection Systems. However, the mechanism of these models is often hard to understand, especially for researchers in the domain of network security. In this paper, we propose a visual analytics system for interpretable deep learning based intrusion detection. During the design of this visual analytics system, we follow the requirements and features of explainable artificial intelligence for users in the domain of network security. The system allows users to select the best parameters to construct the model, to better understand the role of neurons in a deep learning model, to select instances and explore the detection mechanism of the model on these instances. We present multiple use cases to demonstrate the effectiveness of our system.