A Federated Learning-Based Approach for IoT Malware Detection
Houyu Liu, Wanping Liu, Ling Chao Lu · 2025
The rapid iteration of artificial intelligence technology has promoted the large-scale growth of Internet of Things devices, but the common problems such as computing resource constraints and lack of security protection mechanisms in edge devices have led to the continuous expansion of the malware attack surface. The centralized detection method is faced with privacy leakage and resource overload when it is used to prevent Internet of Things threats. Although the traditional distributed architecture can maintain data privacy to a certain extent, its average aggregation mechanism is not robust in the face of Byzantine attacks. To solve the above problems, combined with multi-layer perceptron and dynamic weighted average aggregation algorithm, this paper applies federated learning to the detection of IoT terminal malware, and proposes a new distributed detection method FL-MDwAvg, which fills the theoretical and practical gap of small sample training for IoT terminal devices. Experiments on N-BaIoT public dataset show that compared with other methods, the proposed method achieves a higher accuracy of 99.97% based on small-scale data. Under the full label flip attack, the average accuracy reduction is 4.67% lower than that of the current optimal aggregation algorithm.