Cyber Intrusion Detection Using Shallow Neural Network with Generalized Attack-Sharing Loss
Muhammed Cihat Ünal, Alper Karamanlıoğlu, İsmail Karakaya, Berkan Demirel · 2024
The increasing complexity and frequency of cyber attacks pose significant threats to the security and integrity of computer networks. Machine learning and deep learning algorithms show significant potential in attack detection systems. However, due to the imbalance and diversity of datasets, these models often exhibit a marked bias towards the majority class in many cases. Our paper proposes a generalized attack-sharing loss function to reduce the number of false negatives without damaging the false positives produced by a shallow neural network. Our empirical studies on the CICIDS2017 dataset demonstrate that using our loss function to train a shallow neural network effectively classifies various network attacks.