<scp>DNPFL</scp> : A Federated Deep Learning Framework for Enhanced Network Intrusion Detection With Privacy Preservation

ABSTRACT In the face of increasing cybersecurity threats, effective network intrusion detection systems are crucial for safeguarding sensitive data and maintaining the integrity of digital infrastructures. This research paper presents the DNPFL (Deep Neural Powered Federated Learning) framework, which integrates advanced deep learning techniques with a federated learning paradigm to enhance the detection of various network attacks, including DoS, DDoS, Probe, R2L, and U2R. The proposed method utilizes a hybrid architecture that combines decentralized data processing through federated learning with advanced deep learning techniques, enabling efficient feature selection and classification of various network attack types while ensuring data privacy and security. Utilizing prominent datasets such as NSL‐KDD, CICIDS 2017, UNSW, and DARPA 1998/1999, the model demonstrates remarkable performance metrics, achieving high accuracy. The DNPFL model effectively minimizes false positives while accurately identifying both common and complex attack types. The integration of federated learning ensures data privacy by enabling decentralized model training, while the deep learning architecture captures intricate patterns associated with malicious activities. The outcome highlight the model's robustness and adaptability, pointing it as a valuable solution based on real‐time intrusion detection systems in evolving cyber threat landscapes.