A Defensive Framework Against Metigatic Generic Attacks Targeting Network Imbalance in Machine Learning-Based Cyber Defense Systems

The increasing complexity of cyber threats is making it harder for traditional intrusion detection systems to identify adaptive and subtle attacks. Machine learning (ML)-powered network intrusion detection systems (NIDS) are more capable of detecting threats, but they may be compromised by generic attacks. This can result in network imbalances and poor model performance since hostile traffic is under-represented. In order to improve ML-based NIDS, this research proposes a metigatic defensive architecture. The framework incorporates methods such as adversarial training, dataset balancing, feature preprocessing, feature engineering, and model fine-tuning. We construct realistic adversarial traffic samples with feature interdependencies and protocol compliance to evaluate and enhance the resilience of the model. Tested on the UNSW-NB15 and NSL-KDD datasets, the results demonstrate considerable improvements in accuracy, recall, and precision, particularly for under-represented or adversarial perturbed attacks. According to the results, the proposed architecture mitigates the impact of traffic imbalance brought on by generic attacks, providing a workable and scalable approach for robust intrusion detection in dynamic network environments.