Deep Learning-Assisted Application Traffic Classification for 5G/B5G Network Slicing

The 5G and beyond mobile networks are targeted to offer high data rate, very low latency and ubiquitous connectivity for the end users. These varied Quality of Service (QoS) requirements are attributed to rapid growth of users and numerous applications. To accommodate heterogeneous resource demands and traffic arrival, the Network Slicing (NS) is implemented by the 5G mobile operators. The NS enables logical network isolation between different QoS requirements utilizing the single physical infrastructure. The dynamic resource allocation for each network slice reduces resource wastage and offers uninterrupted connectivity. To efficiently manage the available resources, the precise classification of incoming traffic is required. Further, considering the complexity of traffic flow variations associated with 5G/B5G, the Deep Learning (DL) models are required for traffic classification. Thus, in the proposed work, the DL models are utilized for traffic classification and classified applications are assigned to the respective network slices based on available resources. Around 75 applications were experimented with various deep learning models. Among the models considered, the Gated Recurrent Unit (GRU) is observed to exhibit peak accuracy of 99.5% with comparatively lesser training time consumption.