Efficacy of Spiking Neural Networks for Intrusion Detection Systems

Protection against potential threats is paramount in computer networks and requires robust security measures. However, traditional rule-based Intrusion Detection Systems (IDSs) often fail to adapt to dynamic environments, prompting the exploration of innovative solutions such as Neural Network (NN)-based approaches. Previous advances have primarily focused on conventional NNs. Only more recent studies researched the use of Spiking Neural Networks (SNNs) for IDSs; however, they rely on pre- or post-processing steps in their methods, which interferes with the analysis of the actual applicability of SNNs for IDSs. This study aims to overcome this deficit by analyzing the efficacy of SNNs as the sole data processor for IDSs, i.e., without using any non-essential processing outside of the network (”bare” SNNs). Through extensive experimentation on the NSL-KDD, CIC-IDS-2017, CIC-IOT-2023, and AWID3 datasets, we examined various configurations of bare SNNs, alongside conventional NNs, and Recurrent Neural Networks (RNNs) for comparison. The results demonstrate that SNNs can achieve robust performance for IDSs without the pre- or post-processing steps required by other studies. ... mehrIn detail, the bare SNNs achieved higher or similar accuracy for all datasets compared to the other NN models. Furthermore, a comparative analysis reveals a competitive advantage of SNNs over the other NN models in generating fewer false positives. The results of this study suggest that SNN-based IDSs are a promising direction to strengthen network security. However, further research is essential to ensure broader applicability and scalability.