Federated Learning and Explainable Hybrid Deep Learning for Cyber-Attack Detection in Distributed Big Data Environments: A Comprehensive Survey

The proliferation of distributed computing environments, including cloud computing, IoT networks, and edge systems, has increased the cyber-attack surface and created new challenges in security. Centrally located intrusion detection systems (IDSs) have shortcomings with data privacy, scalability, communication, and transparency. Federated Learning (FL) is a distributed learning system that allows system nodes to collaboratively train a detection model without sharing raw data, providing privacy preservation and increased detection in more nodes. Along with Federated Learning, Explainable Artificial Intelligence (XAI) has been used to improve deep-learning cybersecurity services and the confidence in the services. This survey provides a complete analysis of the combination of federated learning and explainable hybrid deep learning systems used in the detection of cyber-attacks. This is research of the intrusion detection systems based on FL, explainable artificial intelligence systems, privacy-preserving systems, and the machine learning and deep learning systems used in the cybersecurity field. In addition, the challenges of data heterogeneity, costs of communication, privacy leakage, and the explainability gap are examined. The last portion of this survey covers the unfulfilled research needs and the paths this uncharted research can take. This can foster the development of systems to detect cyber-attacks that are explainable, private, and safe in next-generation distributed systems.