Enhancing IoT Intrusion Detection with Federated Learning-Based CNN-GRU and LSTM-GRU Ensembles
Ogobuchi Daniel Okey, Demóstenes Zegarra Rodríguez, João Henrique Kleinschmidt · 2024
The Internet of Things (IoT) devices have been identified to have low security measures by default, thus making them highly vulnerable to malicious attacks. Machine learning-based intrusion detection systems (IDS) are used to mitigate these attacks, however, there is a compromise in security and privacy of data ownership between IoT devices. This paper proposes a Federated Ensemble IDS (CNN-GRU and LSTM-GRU) for monitoring IoT network activities using Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU) and Long-Short Term Memory (LSTM) networks that classifies the network as either normal or malicious. Two aggregation functions, including wFedAvg and wFedProx, are developed to create a global model from different clients’ contribution. We perform an evaluation of the proposed IDS on CICIoT2023 and FLNET2023 datasets. The results show that with wFedAvg, the CNN-GRU achieved an accuracy of 98.25% and 99.25% on the CICIoT2023 and FLNET2023 datasets respectively. Additionally, the LSTM-GRU model shows a detection accuracy of 93.36% and 95.66%, respectively on CICIoT2023 and FLNET2023 datasets. The performance shows that the proposed method is robust enough to enhancing the privacy of IoT devices.