Integrating Machine Learning With Federated Learning for Privacy-Preserving Cyber Threat Detection
Abstract
The high proliferation of distributed computing, IoT platforms, and cloud-edge architecture has compounded the challenge of cyber threats as it has also limited the exchange of sensitive security information. Traditional machine learning-based intrusion detection systems propose centralized data aggregation which is incompatible with privacy laws and data requirements of a company. Although these improvement efforts have been made in the recent times, current federated learning frameworks tend to have a problem with volatile convergence, non-IID data distributions, and lack of resistance to adversarial behavior. This article suggests a unified machine learning-federated learning system to do privacy-aware threat detection on cyber-attacks such that the collaborative model can be trained without the raw security information being exposed. The framework integrates adaptive local learning, robustness sensitive aggregation and secure coordination in enhancing generalization among heterogeneous clients.