Federated Learning for Attack Prediction on UNSW-NB15 Training Data

With the increasing prevalence of cyber threats, robust network security has become imperative. Traditional centralized machine learning models for network intrusion detection systems (NIDS) face significant challenges related to data confidentiality, scalability, and centralization risks. Federated Learning (FL) offers a decentralized approach that allows multiple clients to train a shared model while keeping their data local collaboratively. This technique is particularly relevant to attack prediction in cybersecurity, where organizations may be hesitant to share sensitive data due to privacy and security concerns. This paper investigates the application of FL on the UNSW-NB15 dataset to predict network attacks. The study demonstrates the potential of FL to achieve high accuracy and minimal false negatives in attack detection while preserving data confidentiality. The results visualized in the confusion matrix highlight the effectiveness of FL in distinguishing between normal and malicious network traffic, making it a promising approach for real-world cybersecurity applications. By leveraging FL, organizations can improve their network security infrastructure while reducing the risks associated with centralized data processing.