Deep Convolutional Neural Network for Improving Intrusion Detection. A Spectogram based NIDS Framework

In the past few years, computer networks have changed drastically as the number of connected devices and applications increased significantly. Because of the network’s increased scale and our reliance on it for all parts of our lives, hostile parties have launched a slew of new attacks against the network, as well as mutations of existing attacks. When it comes to safeguarding devices and network nodes, along with the data they contain, against possible breaches, these attacks provide a number of challenges for network security professionals. One of the best security solutions is a system to detect network intrusions which continuously monitors network traffic to safeguard network entry points. When it comes to identifying new assaults, the false alarm rate (FAR) for NIDS remains high even after substantial scientific efforts. We introduce a novel architecture for Network Intrusion Detection Systems (NIDS) employing a deep convolutional neural network alongside spectrogram images generated through the short-time Fourier transform. The effectiveness of our approach was assessed using the CIC-IDS2018 dataset. Our experimental findings revealed a reduction in False Alarm Rate (FAR) by 4.3