Advanced Cybersecurity Framework for Intrusion Detection Utilizing Federated Machine Learning

Abstract

Cybersecurity risks have grown a lot as the number of devices connected to the Internet of Things (IoT) has proliferated. It's hard for old breach detection systems to keep up with new threats and protect data protection at the same time. This study suggests a way to find intrusions that don't invade privacy by using a Deep Neural Network (DNN) built on Autoencoders and Federated Learning (FL). Our method lets many devices work together to train the model without sharing private information, improving security and privacy. We test the model using standard datasets like RT-IoT 2022, CIC IoT 2023, BoT-IoT, IoT-23, and CIC IoMT 2024. We were able to show that our model can correctly find large-scale threats like DDoS, DoS, DEnial-of-Service, and malware. But it's still hard to spot sneaky attacks like Man-in-the-Middle (MITM), Spoofing, and Command & Control (C&C) because memory rates are so low. Also, FL makes data safer, but it adds extra work to computation and communication, which makes realtime distribution harder. Our results show how essential dataset properties, feature selection, and hyperparameter tuning are for making models work better. Future studies should focus on developing FL to work better with big IoT networks and finding more advanced hacking dangers. Even though there are some problems, our method is a good step toward strong IoT security that protects user privacy.