A Multi-Layered Representation for Intrusion Detection System in Cyber Systems Using CNN Deep Learning Algorithm

Convolutional Neural Networks (CNNs), a kind of deep learning architecture, have recently shown remarkable progress in the field of computer vision. This study models network traffic as a time-series, in particular transmission control protocol/internet protocols (TCP/IP) packets in a predetermined time range, and applies supervised learning techniques to improve performance for the job of detection of intrusions (ID) in cyber security. Each experiment uses a learning rate between [0.01] and [0.05] to run the models for up to 1000 epochs. When compared to traditional machine learning classifiers, CNN and its different architectures perform very well. This is primarily because CNNs can abstractly represent low-level characteristic sets of traffic on network links in the form of high-level features.