FedSecure-IoT: Privacy-Preserving Intrusion Detection for IoT Networks Using Hybrid Deep Learning, Federated Learning and Homomorphic Encryption

Abstract

FedSecure-IoT is a privacy-preserving intrusion detection framework for IoT networks. It combines a hybrid CNN-LSTM-DNN classifier, federated learning with Federated Averaging (FedAvg), and Paillier homomorphic encryption for protecting model updates during aggregation. On traffic derived from the CICIoT2023 dataset (8,000-sample balanced test set), a centralized baseline reaches 99.22% accuracy and plain federated learning reaches 97.52%. The encrypted federated variant currently reaches only 40.20% accuracy, and the report discusses candidate causes and next steps. Code: https://github.com/NitinRajvanshi/FedSecure_IoT