D-DSQN: A Coordinated Botnet Suppression Mechanism for Social IoT Networks

The Internet of Things (IoT) with social relationships has fundamentally transformed communication between various smart devices, while this transformation poses significant challenges for the propagation of botnets. To address this issue, we propose an optimal strategy for controlling botnet propagation in a social IoT environment. Specifically, we introduce a botnet propagation framework that incorporates social participant factors and conduct an in-depth analysis of the role and impact of social participants in the botnet propagation process. Subsequently, we construct a botnet propagation control model and formulate an optimization problem based on this model using a partially observable Markov decision process (POMDP). To determine the optimal control strategy in practice, we propose dueling-doubly spiking Q -network (D-DSQN), a deep reinforcement learning (DRL)-based botnet propagation control algorithm that integrates spiking neural networks (SNNs) into the dueling double deep Q network (D3QN) algorithm to enhance learning performance. Furthermore, we discuss the sensitivity of D-DSQN to hyperparameters and demonstrate its stability through experiments. Finally, extensive experiments are conducted in both simulation-based social IoT environments with varying complexity and a real-world IoT botnet traffic dataset. The results show that D-DSQN achieves faster learning convergence, stronger robustness, and higher cumulative rewards in dynamic social IoT scenarios, while also delivering superior detection performance on real network traffic.