Deep Attention Learning for Extreme Minority Class Intrusion Detection in Network Traffic

In the expansive realm of the Internet, escalating online traffic corresponds to a surge in sophisticated network attacks. Intrusion Detection Systems (IDS) are pivotal in identifying these threats, with deep learning neural networks proving effective in processing extensive data. However, imbalanced data in cybersecurity poses a challenge, hindering the accurate detection of minority attack classes. This study utilizes a Deep Neural Network for intrusion detection, exploring variations in parameters and focusing on minority classes in imbalanced multi-class data. Experiments on the CICIDS-2017 dataset reveal that certain coarse-grained features play a crucial role, enabling accurate detection even with minimal instances. This underscores the significance of specific feature characteristics in identifying minority class threats within the dynamic landscape of cybersecurity.