An Improved Machine Learning Approach for Throughput Prediction in the Next Generation Wireless Networks

Currently evolving 5G telecom networks require different intelligent learning and decision mechanisms to adapt to the varying network conditions. Further, considering additional requirements of low-latency and ultra-reliability, newer resource allocation schemes are to be explored to find the most effective way to predict the amount of resources required by a system. In this paper, a unique resource allocation scheme is devised for the 5G network using the properties of first packet transmission and the subsequent retransmissions. The proposed model is shown to accurately predict the system-level as well as user-level throughput for a set of mobile users, while ensuring lower consumption of network resources. The simulations show an accuracy of 85% for the user-level throughput that is acceptable for the dynamic resource planning in future networks.