Exploiting Machine Learning for the Performance Analysis of a Mobile Hotspot with a Call Admission Control Mechanism

Abstract

Machine Learning (ML) algorithms can be efficiently employed to calculate various performance metrics in telecommunication systems showing comparable accuracy with analytical expressions while at the same time decreasing the computation time in several operational cases. In this paper, we examine the impact of six ML methods both on the accuracy of calculations and on the estimation time and benchmark them against an analytical formalism which solves a 2D Markov chain to estimate seven performance metrics in a vehicular system of a mobile hotspot. As a consequence, when using ML methods, we show that the computational complexity can be reduced, especially in cases where the system capacity is large and the computational complexity of the 2D Markov chain increases. More specifically, the proposed approach is applied in a dataset which comprises 100,000 operational cases, demonstrating a reduction of estimation time of more than two orders of magnitude while maintaining the average error less than 4.5%.