The upcoming mobile core network (5G) is expected to support Enhanced Mobile Broadband, Massive Machine Type Communication (MTC) and Ultra-low latency, within the same infrastructure. As those services categories have different expectations in terms of network connectivity, QoS and Key Performance Indicators, some enhancements are being planned to make the network more flexible through the integration of new technologies like Network Functions Virtualization (NFV) and Software Defined Networks (SDN). Following our benchmark conducted in a previous work , the Mobility Management Entity (MME) newly known as Access and Mobility Function (AMF) in the New Generation Core (NGC) architecture, is often overloaded due to the fact of being the only access component over the control plane for a huge number of connected devices. Therefore, latency is increased over not only the attach requests but also over other procedures such as authentication or handover.. At IRT b<>com, a complete strategy is being developed using techniques originating from control theory to tackle the issues introduced above. The strategy consists on dispatching the attach request or any mobile network procedure based on the load of the AMFs deployed in the NGC. Therefore, new arrivals are treated with no additional latency. Furthermore, based on control theory, the strategy can scale in or out the AMF depending on the increase or the decrease of the arrival load. In this poster we are going further by using new advances on machine learning, and more specifically Recurrent Neural Networks (RNN), to predict accurately the arrival traffic pattern of devices. The main objective of the proposed approach is to early react to congestion by pro-actively scaling the AMF VNF in a way to absorb such congestion while respecting the traffic constraints.