FEDERATED HYBRID DEEP LEARNING FOR NETWORK ANOMALY DETECTION WITH ADAPTIVE RESOURCE OPTIMIZATION

Abstract

The rapid growth of distributed systems and networked environments has increased the complexity of real-time traffic analysis and management. Traditional centralized approaches face limitations related to latency, scalability, and data privacy. This study proposes a federated hybrid deep learning framework for network anomaly detection combined with adaptive resource optimization. The model integrates Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Autoencoders to capture spatial, temporal, and reconstruction-based anomaly patterns. A federated learning strategy using Federated Averaging (FedAvg) enables decentralized training across multiple edge devices with non-IID data distribution, preserving data privacy. Additionally, a Deep Q-Network (DQN) is employed to dynamically optimize network resource allocation based on detected anomalies and traffic conditions. The framework is evaluated using the UNSW-NB15 dataset and compared with traditional machine learning models and centralized deep learning approaches. Results demonstrate improved detection performance and efficient resource utilization, making the proposed system suitable for real-world distributed network environments.