  Citation: Tomic, I.; Bleakley, E.; Ivanis, P. Predictive Capacity Planning for Mobile Networks—ML Supported Prediction of Network Performance and User Experience Evolution. Electronics 2022, 11, 626. https://doi.org/10.3390/ electronics11040626 Academic Editors: Jihoon Yang and Unsang Park Received: 31 January 2022 Accepted: 15 February 2022 Published: 17 February 2022 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affil- iations. Copyright: © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). electronics Article Predictive Capacity Planning for Mobile Networks—ML Supported Prediction of Network Performance and User Experience Evolution Igor Tomic 1,2, *, Eoin Bleakley 3 and Predrag Ivanis 1 1 School of Electrical Engineering, University of Belgrade, 11070 Belgrade, Serbia; predrag.ivanis@etf.bg.ac.rs 2 Aspire Technology Unlimited, 11000 Belgrade, Serbia 3 UCD School of Electrical and Electronic Engineering, University College Dublin, D04 V1W8 Dublin, Ireland; eoin.bleakley@gmail.com * Correspondence: igor.tomic@aspiretechnology.com Abstract: Network performance prediction is crucial for enabling agile capacity planning in mobile networks. One of the key problems is predicting evolution of spectral efficiency in growing network load conditions. The main factor driving network performance and spectral efficiency is reportedly the Channel Quality Indicator (CQI). In this paper, the performance of different Machine Learning (ML) models were examined, and XGBoost was selected as the best performing model. Furthermore, to improve modeling accuracy, several features were introduced (operating frequency band, Physical Resource Block (PRB) utilization in surrounding cells, number of surrounding cells within a radius, heavy data factor and higher order modulation usage). The impact of these features on CQI prediction were examined. Keywords: mobile network; 5G-NR; LTE; user experience; performance modeling; prediction; CQI; Multiple Input Multiple Output (MIMO); network load; XGBoost 1. Introduction Capacity planning for mobile networks has been a challenge for network planners over the past decade. Traffic in mobile networks has grown exponentially. The growth rate has varied by market, but on average traffic has doubled every two years [1]. This pace of development mirrors Moore’s Law. In parallel with growing load, network performance dynamically changes, with expected performance downgrade [2] in cases where there is a lack of investments in additional capacity. Simultaneously, user demand, or more precisely application demand, for throughput and latency has only increased. Further complexity is added by the constraint that the process of adding capacity to mobile networks comes with long cycles. Mobile operators typically need six months to add a 4G or a 5G layer, and two years to build a new base station. Finally, there is always huge pressure to justify all Capital Expenditure (CapEx) investments. In such circumstances, predictive planning is a must. The decision-making process on capacity addition needs to be based on the accurate estimation of future network performance, and what-if evaluations of different scenarios of traffic growth, network performance and capacity expansions. The problem of predicting user experience in terms of data throughput in mobile networks of the fourth and fifth generation (4G and 5G), based on Orthogonal Frequency Division Multiple Access (OFDMA) techniques, can be decoupled into two parallel streams. Spectrum assets of typical mobile operators are spread over channels in different frequency bands. Channel bandwidth in Long Term Evolution (LTE) systems is 5, 10, 15 or 20 MHz, while in 5G it can be 50–100 MHz in lower frequency bands, and up to 400 MHz on higher frequency bands. Both LTE and 5G systems have resource grids deployed over channels, where the available spectrum is split into Resource Blocks (RBs). In LTE each RB has a Electronics 2022, 11, 626. https://doi.org/10.3390/electronics11040626 https://www.mdpi.com/journal/electronics