A Deep Learning Framework for IoT Lightweight Traffic Multi-Classification: Smart-Cities

Increased traffic volume is a major challenge for effective network management in the wake of the proliferation of mobile computing and the Internet of Things (IoT). Earlier models surrender efficiency to achieve high-precision classification outcomes, which are no longer fitting for limited assets in edge network circumstances, making traffic classification a difficult task for network administrators everywhere. Given the nature of the problem, the current state of the art in traffic classification is characterized by extremely high computational complexity and large parameters. Increased traffic volume is a major challenge for effective network management in the wake of the proliferation of mobile computing and the Internet-of-Things(IoT). Earlier models surrender efficiency to achieve high-precision classification outcomes, which are no longer fitting for limited assets in an edge network circumstances, making traffic classification a difficult task for network administrators everywhere. Given the nature of the problem, the current state of the art in traffic classification is characterized by extremely high computational complexity and large parameters. To strike a clever balance between performance and size, we present a deep learning (DL)-based traffic classification model. We begin by decreasing the amount of model parameters and calculations by modifying the model's scale, width, and resolution. To further improve the capability of feature extraction at the traffic flow level, we secondly incorporate accurate geographical information on the attention mechanism. Thirdly, we get multiscale flow-level features by employing lightweight multiscale feature fusion. To strike a clever balance between performance and size, we present a deep learning (DL)-based traffic classification model. We begin by decreasing the amount of model parameters and calculations by modifying the model's scale, width, and resolution. To further improve the capability of feature extraction at the traffic flow level, we secondly incorporate accurate geographical information on the attention mechanism. Thirdly, we get multiscale flow-level features by employing lightweight multiscale feature fusion. The results of our experiments demonstrate that our model has high classification accuracy and efficient operation. Our study presents a traffic categorization model with an accuracy of over 99.82%, a parameter reduction of 0.26M, and a computation reduction of 5.26M. Therefore, this work offers a practical design used in a genuine IoT situation, where IoT traffic and tools' profiles are anticipated and classified while easing the data dispensation in the higher levels of an end-to-end communication strategy.