Research On Visual Analytics System For Network Security Detection Algorithm

Network security is becoming increasingly serious due to the complex network environment.Cyber-attacks will not only raise privacy violation issues,but also cause enormous financial loss.Faced with large-scale network data,deep learning has become an effective intrusion detection method.We train deep learning models on large-scale data-set and the model then learns the features of normal and attack data to classify them.Deep learning models demonstrate promising performance on network intrusion detection problems.We encoded and transformed network security data into the form of two-dimensional gray-scaling images.We then used a convolutional neural network to classify the transformed images into attack or normal class.The existing network security data often presents the class imbalance problem.We may only have a small amount of data for certain attack types,which makes the deep learning models hard to learn the features of such attacks.We proposed a GAN-based data augmentation method to generate new data of such types and add them into the original training set to solve the class imbalance issue.We demonstrated the effectiveness of our method compared with existing methods on solving the class imbalance problem through experiments.Our method shows more increase in the accuracy of the intrusion detection model.Although deep learning models have achieved satisfying numerical results in network intrusion detection problems,it is hard for users in the field of network security to understand the working mechanism of such models,since they learn abstract high-level representation of data to make the classification.How to choose a proper deep learning model structure,understand the decisions made by the model and then trust the model is what we need to solve to increase the real application of deep learning models in the network security field.We proposed a visual analytics system for network security detection algorithm.We designed a model selection module and a feature analysis module to help our users to compare the performance of different deep learning models and understand what features the model use to make the classification.We designed several use cases to demonstrate the effectiveness of our visual analytics systems.