Transformer-Based Feature Extraction for Intrusion Detection in Cloud-Native Environments

Cloud-native environments require intelligent and scalable intrusion detection mechanisms to defend against increasingly sophisticated cyber threats. This paper proposes a Transformer-assisted intrusion detection framework that integrates the CICIDS2017 and UNSW-NB15 datasets through feature harmonization. After preprocessing and feature selection, a Transformer encoder learns contextual feature embeddings, which are combined with the selected features to form a hybrid representation. The proposed framework combines contextual feature learning with gradient boosting to improve the effectiveness and scalability of intrusion detection in cloud-native environments.