Attention Mechanisms and Graph Neural Networks for IoT Botnet Detection: A Comprehensive Survey, Taxonomy, and Future Research Directions PREDICTION

There is a massive increase in the number of IoT devices connected to the Internet; this large number of IoT devices creates a vast amount of potential vulnerabilities creating an attack vector and making it difficult to prevent or mitigate Botnets and DDoS Attacks. With most IoT infrastructure transitioning from cloud based architectures to Edge/Fog based architectures many traditional centralized security architectures are no longer feasible due to the complexity and heterogeneity of these networks as well as their limited resources. Therefore, this research paper will be the first of its kind to provide an exhaustive comparison of the various methodologies used to detect botnets. It will examine the evolutionary progression from traditional ML, hybrid CNN-LSTM models and then to current state-of-the-art GNN models with a focus on GNN models using fast learnable multi-attention mechanisms to detect botnets. Using a thorough literature review of all published studies, it will show how GNN-based models have achieved better results than traditional methods (99.2-99.3%) but still have significant deficiencies in detecting zero day attacks, and leaking user information during the process of constructing graphs and adapting to changes in data concepts. Finally, this paper will synthesize these findings into a road map for future research by identifying that there is a need to integrate FL and DP into GNIIDS.