Abstract
The proliferation of Internet of Things (IoT) devices in smart cities has revolutionized urban infrastructure while escalating the risk of botnet attacks that threaten essential services and public safety. This research addresses the critical need for intrusion detection and mitigation systems by introducing a novel hybrid deep learning model, Stacked Autoencoder-Gated Recurrent Unit (SAE-GRU), specifically designed for IoT networks in smart cities. The study targets the dual challenges of processing high-dimensional data and recognizing temporal patterns to identify and mitigate botnet activities in real time. The methodology integrates Stacked Autoencoders for reducing dimensionality and gated recurrent units for analyzing sequential data to ensure both accuracy and efficiency. An emulated smart city environment with diverse IoT devices and communication protocols provided a realistic testbed for evaluating the model. Results demonstrate significant improvements in detection performance with an average accuracy of 98.65 percent and consistently high precision and recall values. These findings enhance the understanding of IoT security by offering a scalable and resource-efficient solution for botnet detection. The functional investigation establishes a foundation for future research into adaptive security mechanisms that address emerging threats and highlights the practical potential of advanced deep learning techniques in safeguarding next-generation smart city ecosystems.