A Federated Learning Based Deep Autoencoder Framework for Robust Malware Detection in Edge Cloud Networks

Abstract

: The edge cloud-based Internet of Things (IoT) devices are diverse in size, type, and the function it performs. The tremendous increase in the number of cyberattacks, coupled with the adverse technologies, has rendered current security measures increasingly ineffective. Addressing security and privacy issues in edge cloud networks using machine learning and deep learning-based solutions is effective up to a certain extent; however, these intrusion detection solutions (IDS) rely heavily on the quantity and quality of data. This dependency results in high inaccuracy and high false positive rates in identifying malware. Moreover, the existing machine learning-based malware detection techniques are less efficient and fail to detect malware in unknown or newly joining edge cloud IoT devices in the network. In this paper, our contribution is twofold: Firstly, a federated learning-based method using a deep autoencoder (DAE) has been proposed to detect malware attacks in the edge cloud network. This approach enhances privacy by ensuring that the device data is not transmitted or moved outside the network edge. Instead, the deep learning computation occurs where the data is generated, providing better data protection. Secondly, we have designed a novel method that utilizes DAE, which can effectively detect malware in unknown or new IoT devices. The experimental results show that the proposed DAE model has achieved an AUC of 89%. Similarly, we have compared the performance of our proposed model with other deep learning models such as DNN, CNN, and RNN, and the proposed model has 5% better accuracy than CNN, 17% better than DNN, and 21% better accuracy than RNN in detecting malware in both known and unknown devices.