Evolutionary Convolutional Neural Network: An Application to Intrusion Detection

Abstract

Intrusion detection system (IDS) plays a significant role to secure our privacy data, which can avoid various threats from Internet. There are more and more research studies to use convolutional neural networks (CNNs) as IDSs. However, it is still very challenging on how to develop a reliable and effective IDS by using CNNs. Thus, this paper suggests an evolutionary convolutional neural network (ECNN) as an IDS. It is a first try to use multiobjective immune algorithm to simultaneously optimize the accuracy and weight parameters of CNNs. Such that, our method can obtain various CNN models with different detection accuracies and complexities. The users can select their preferences based on their security requirements and hardware conditions. A number of experiments have been conducted on the NSL-KDD and UNSW-NB datasets to study the capability and performance of the proposed method. When compared to some state-of-the-art algorithms, the experimental results show that our method can obtain a higher detection accuracy.