FedFusion-IDS is a new privacy-preserving federated framework for intrusion detection based on federated learning, multi-objective feature selection with NSGA-II and SHAP second-level filter, hybrid deep learning architecture with CNN, BiLSTM, and Attention mechanisms, explainability through SHAP and LIME, and zero-day anomalies detection.
Abstract
Network Intrusion Detection Systems (NIDS) are vital for protecting cybersecurity infrastructures. The existing intrusion detection methodologies that rely on centralization do not offer privacy protection, cannot protect from zero-day attacks, and lack transparency needed to comply with modern cybersecurity policies. In this work, we present FedFusion-IDS, a new privacy-preserving federated framework for intrusion detection. It is based on federated learning, multi-objective feature selection with NSGA-II and SHAP second-level filter, hybrid deep learning architecture with CNN, BiLSTM, and Attention mechanisms, explainability through SHAP and LIME, and zero-day anomalies detection. We perform a thorough evaluation of FedFusion-IDS on the modern TON_IoT dataset with 10 federated clients for 20 communication rounds. During the multi-objective optimization phase, a Pareto-optimal feature subset is selected to decrease dimensionality while retaining important traffic information. The local clients train a CNN-BiLSTM-Attention model with 962,181 parameters, and parameter aggregation is done with conventional FedAvg and F1-weighted adaptive aggregation techniques. From the experiments performed using TON_IoT dataset, global detection accuracy was found to be 66.26%, F1-score was calculated as 0.6524, and ROC-AUC value exceeds 0.99 for all primary attacks (Normal: 0.999, Backdoor: 1.000, DDoS: 0.997, DoS: 0.700, Password.Dual SHAP and LIME explainability mechanisms deliver fine-grained global feature rankings and instance-level explanations, ensuring compliance with legal transparency requirements such as GDPR Article 22 and the EU AI Act. Furthermore, zero-day anomaly detection simulations confirm that FedFusion-IDS effectively identifies out-of-distribution threats without prior signature training. Together, these results provide a practical blueprint for deploying privacy-preserving, explainable intrusion detection across distributed IoT networks.
The proposed framework introduces several innovative features, such as federated learning with momentum-based optimization, adaptive differential privacy, trust verification via blockchain, and Byzantine-resilient aggregation, to enhance the security, scalability, and robustness of the system compared with traditional...
A Federated Temporal-Aware Lightweight IDS (FTL-IDS) that co-designs multi-domain CAN feature extraction, an LSTM–Transformer temporal encoder, 8-bit quantization-aware training (QAT), and federated learning (FL).
Hamed Alqahtani, Gulshan Kumar· Multimedia tools and applica...· 0 citations
PPFL-IDS combines federated model aggregation with differential privacy noise injection and secure aggregation protocols to train a lightweight gradient-boosted ensemble IDS without exposing local device data, demonstrating that strong privacy guarantees and high detection accuracy can be achieved simultaneously in fed...
Nutan Gusain, J. Alzubi· International Journal on Com...· 0 citations
The results show a success in implementing a real time, scalable, privacy-preserving, and adaptive IDS in large-scale IoT deployments through intelligent workload distribution between edge and cloud layers.
Chidera Winifred John, Eduediuyai Ekerete Dan, P. Asuquo et al.· E3S Web of Conferences· 0 citations
Zero-day is a type of attack that targets vulnerabilities unknown to vendors and security experts. Traditional signature-based intrusion detection systems fail to flag such attacks. ML models used to detect attacks require data aggregation, which raises substantial privacy concerns and infringes on an organization’s da...
Gargi Chaudhari, Zenia Nouphal, S. Vengurlekar· International Conference Inn...· 0 citations
Evaluated on standard intrusion detection datasets, including NSL-KDD and CICIDS2017, indicates that the proposed model can maintain competitive detection performance while improving privacy protection and supporting deployment across distributed network environments.
Ram Chouhan, Kirti Jain, M. Bagwani· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 9, 2026