Network Intrusion Detection using Dynamic Weighted Aggregation Federated Learning

Cybersecurity threats, pervasive in our daily activities due to network openness, are exemplified by malicious exploitation of vulnerabilities. These illicit activities, including unauthorized data alterations, emphasize the urgency for effective defense. Integrating Federated Learning (FL) insights from diverse sources presents a compelling approach, addressing network intrusion limitations and heralding a more resilient security paradigm. In this study, we proposed an efficient FL system for network intrusion detection, integrating a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) based learning model. The main goal of this research is to enhance intrusion detection while upholding data privacy using FL. We conduct experiments using different numbers of clients (5, 10, and 15) and apply the Dynamic Weighted Aggregation Federated Learning (DWAFL) technique to collaboratively train the model across the clients data. DWAFL is basically an approach where model aggregation incorporates dynamic weightings based on the performance of individual client’s model. The experimental results demonstrate that the proposed system achieves 92.2% accuracy with 5 clients, 94.2% with 10 clients and 93.2% with 15 clients, using DWAFL technique. These findings showcase the potential of FL with DWAFL in intrusion detection scenarios, allowing accurate modeling with distributed data sources while preserving data confidentiality. The proposed approach contributes to the development of collaborative learning systems for intrusion detection applications.