Machine Learning Analysis of Multi-Radio Access Technology Selection in 5G NSA Network

Aug 5, 2026

Upload by Logic

Machine learning supports radio access technology decisions as 5G deployments rely on underlying 4G LTE in both stand-alone and non-stand-alone architectures. RAT selection between 4G LTE and 5G NR must account for geography, mobility, and coverage. The study records live radio measurements over 300 meters between a pedestrian user and a 5G NSA base station, formulates RAT selection as classification, and trains DT, XTREE, RF, GB, and XGBoost to choose the appropriate RAT and enable efficient resource allocation. XGBoost achieves the highest accuracy (93.86%).

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Introduction

5G NSA vs SA and Multi-RAT context

Motivation for RAT selection

Factors guiding RAT selection

Machine Learning Approach

Classification problem formulation

Selected ML algorithms

Evaluation and Results

Classification metrics and accuracy comparison

Best-performing model for RAT selection

Machine Learning Analysis of Multi-Radio Access Technology Selection in 5G NSA Network Nurudeen Oladehinbo Salau and Muhammad Zeeshan Shakir Abstract—The exponential growth of trafﬁc across the mobile networks called for exploitation of new spectrum bands of 5G networks; whose deployment still rely on support from underlying 4G long term evolution (LTE) networks in both stand-alone (SA) and non-stand-alone (NSA) architectures. This scenario poses challenges on the choice of Radio Access Technology (RAT) selection between 4G LTE and 5G new radio (NR) networks to these ever increasing mobile users with respect to their geographical location, mobility and network coverage. Hence, this study investigates joint user requirements and network constraints for appropriate RAT selection between 4G LTE and 5G NR by recording live radio measurements over a distance of 300 meters between a pedestrian user and 5G NSA base-station. The problem (RAT selection) was formulated as a classiﬁcation process, hence implemented with classiﬁcation machine learning (ML) algorithms: Decision Tree (DT), Extra Tree (XTREE), Random Forest (RF), Gradient Boosting (GB), and eXtreme Gradient Boosting (XGBoost) to select an appropriate RAT. Evaluation of results with standard classiﬁcation metrics, show measure of accuracy of algorithms: DT at 91.82%, RF at 87.64%, XTREE at 86.75%, GB at 91.86%, and XGBoost at 93.86%, where XGBoost showed highest performance value, therefore proposed as ML model for RAT selection to achieve effective and efﬁcient resource allocation. Index Terms-5G, Multi-RAT, SA, NSA, Machine learning. I. INTRODUCTION The tremendous trafﬁc increase in mobile networks especially with current usage of smart phones and wireless devices such as tablets and laptops operating between 700MHz and 6GHz with channel bandwidth of 5 to 100MHz [1], resulted into spectrum crunch, in order to meet up with this challenge; the 5G networks are explored coupled with support from underlying network 4G; long term evolution (LTE) which has been tested and trusted, in both stand-alone (SA) and non stand-alone (NSA) architectures. Practically, NSA combines multiple radio access technologies by interwiring with existing 4G LTE network [2]; i.e., the 5G NR radio cells are combined with 4G LTE radio cells using dual connectivity (DC) to provide radio access before connecting to the core network (CN); which can either be evolved packet core (EPC) or 5G core (5GC), to offer higher mobile broadband speeds through its enhanced mobile broadband (eMBB) features. Presently, existing 5G deployments are based on NSA architecture which still require support of 4G LTE infrastructure, however, 5G SA which is a pure end-to-end 5G network uses only one RAT to connect to the 5GC core network, that is the 5G NR radio cells are used for both control plane and user plane. Although 5G SA N. O. Salau and M. Z. Shakir are with the School of Computing, Engineering and Physical Sciences, University of the West of Scotland, Paisley, Scot land, [UK. Email: Nurudeen.Salau@uws.ac.uk](UK. Email: Nurudeen.Salau@uws.ac.uk), [Muhammad.Shakir@uws.ac.uk](Muhammad.Shakir@uws.ac.uk) architecture will no longer depend on 4G infrastructure but there will still be inter-generation handover between 4G and 5G [2] for service continuity, i.e., we will have 4G SA and 5G SA in multiple radio access technology (Multi-RAT) . This will stimulate seamless connectivity of users for ultra-reliable low latency communication (uRLLC) experience, network slicing capabilities and massive machine-type communication (mMTC) applications for better support to internet of things (IoT) devices. Therefore, in the present and nearest future mobile networks deployment, 5G will co-exist with different technologies like 4G in a Multi-RAT environment [3], [4], hence our main motivation of this study of RAT selection between 4G LTE and 5G NR RATs. Fundamentally, the term RAT simply connotes the type of cellular netwo

FAQs

What problem does the study address in 5G NSA networks?

It addresses choosing the appropriate Radio Access Technology between 4G LTE and 5G NR while 5G deployments depend on underlying 4G LTE in NSA architectures.

How is the RAT selection decision modeled in the study?

The RAT selection problem is formulated as a classification process and implemented using machine learning algorithms.

Which machine learning algorithm performs best for RAT selection and what accuracy is reported?

XGBoost shows the highest performance with an accuracy of 93.86%.

Aug 5, 2026

Upload by Logic