Feature Representation for Network Intrusion Detection System Trough Embedding Neural Network

Abstract

Computer network technology is growing rapidly, but cyberattacks are also increasing in number and variants that occur every year. Anomaly-based network intrusion detection system (NIDS) is still vulnerable to false positive rates even though it has used a machine learning approach to detect zero-day attacks on network traffic. Deep learning can provide advanced solutions to this problem. However, deep learning requires special handling to process tabular NIDS datasets with highly sparse categorical and numerical data. To overcome this, we propose a new embedding method implemented by embedding not only categorical data but also numerical data to provide the best representation features for deep learning models. The proposed method was evaluated with other deep learning and machine learning models with results outperforming all models based on the f1-score macro using the CSE-CIC-IDS-2018 dataset.