Enhancing Network Security Through Deep Learning: Overcoming Feature Engineering Limitations

Abstract

The significant impact of cyberattacks on their targets, security domains, particularly those integrated with machine learning (ML), has gained attention in building robust and secure networks. Malware detection is of paramount importance in network security, with recent years presenting a challenge to create Intrusion Detection Systems (IDS) capable of accurately detecting and classifying malware traffic based on raw network data; without involving collecting data, labels, and feature extraction to achieve high accuracy while minimizing false positives. In this proposal, we introduce a Deep Learning model designed to offer a robust system that can detect and classify malware traffic based on a CNN Model and using raw flows of traffic. It's worth mentioning that the raw flows, obtained directly from the monitored stream of bytes, serves as an input feature for the proposed model, without the need for any handcrafted features. This approach aims to achieve high accuracy and reduce false positives to mitigate intrusions and cyber threats effectively.