Double DQN-GAMO: A Cyber Threat Detection Framework for Zero-Day Attacks

To address the growing threats of Zero-Day attacks, we propose an advanced intrusion detection system framework that integrates a GAMO model for data balancing and a Double DQN mechanism for dynamic sample selection. Unlike existing methods relying on static thresholds, our framework continuously adjusts its sampling strategies based on changing threat landscapes, thereby enhancing Zero-Day attack detection and improving the recognition of minority classes in imbalanced datasets. We validate the proposed framework using the CICIDS2017 dataset. In binary classification experiments, our approach achieves 99.66% accuracy, outperforming multiple baseline models. The GAMO component enhances the detection rates for minority attack classes.