Network Intrusion Detection Based On Deep Learning Under Data Imbalance

Industrial Io T is the most rapidly growing industry in the current Io T industry,and intrusion detection systems remain one of the key technologies for industrial Io T security protection.Researchers and scholars consider applying algorithms such as machine learning and deep learning to network intrusion detection systems to cope with complex and variable network environments and to automatically extract key features from high-dimensional feature data.However,in realistic industrial Io T environments,data imbalance is the main factor affecting the performance of deep learning-based intrusion detection systems.In this paper,we will study network intrusion detection techniques under sample imbalance,build a network intrusion detection system using deep learning methods,and construct simulation experiments for analysis based on CSE-CIC-IDS2018 dataset.The main research contents are as follows.Firstly,this paper summarizes and generalizes the theories related to intrusion detection systems under data imbalance.The basic theoretical framework of intrusion detection systems is analyzed.This paper summarizes the research status of machine learning-based intrusion detection,deep learning-based intrusion detection,and intrusion detection under data imbalance.In this paper,the intrusion detection systems are classified according to four criteria.We describe the characteristics of several public datasets for intrusion detection and the CSE-CIC-IDS2018 dataset used in this paper is introduced in detail.This paper investigates the general solution algorithms under data imbalance.Secondly,this paper investigates the deep learning based intrusion detection system.In this paper,convolutional neural networks are introduced.We convert the one-dimensional format of intrusion detection data into a two-dimensional grayscale graph format and input it to the convolutional neural network-based intrusion detection system.This paper also introduces recurrent neural network and its variant model,and constructs an intrusion detection system based on the recurrent neural network variant.Subsequently,considering the low detection accuracy of few classes,we construct an intrusion detection system based on multi-model joint decision making.Experiments are constructed and analyzed based on the CSE-CIC-IDS2018 dataset.The IDS detection accuracy,false alarm leakage,system real-time and class-specific detection are compared and analyzed.The experimental results demonstrate the effectiveness of the intrusion detection system based on multi-model co-decision and find that the model can reduce the model leakage rate.Finally,this paper investigates a network intrusion detection model based on the data level.The principles of autoencoder,variational autoencoder and conditional variational autoencoder are introduced,and this paper constructs three data-based research schemes,namely,a data enhancement scheme based on variational self-encoder,a data balancing scheme based on conditional variational autoencoder,and a data balancing scheme based on random under-sampling and conditional variational autoencoder.The three data-based level schemes are combined with a deep learning-based intrusion detection system,and this paper constructs experiments based on the CSE-CIC-IDS2018 dataset to verify the effectiveness of the three data processing schemes.