Towards ultra-latency using deep learning in 5G Network Slicing applying Approximate k-Nearest Neighbor Graph Construction
Abstract The 5G Network Slicing with SDN and NFV have expended to support new-verticals such as intelligent transport, industrial automation, remote healthcare. Network slice is intended as parameter configurations and a collection of logical network functions to support particular service requirements. The network slicing resource allocation and prediction in 5G networks is carried out using network Key Performance Indicators (KPIs) from the connection request made by the devices on joining the network. We explore derived features as the network non-KPI parameters using the k-Nearest Neighbor (kNN) graph construction. In this paper, we use kNN graph construction algorithms to augment the dataset with triangle count and cluster coecient properties for ecient and reliable network slice. We used deep learning neural network model to simulate our results with KPIs and KPIs with non-KPI parameters. Our novel approach found that at k=3 and k=4 of the kNN graph construction gives better results and overall accuracy is imroved around 29%.