Page 19 - Proceedings of the 2018 ITU Kaleidoscope

IMPACT OF MACHINE LEARNING IN 5G PLANNING
AND DEPLOYMENT

Hugo Miguel

Under Secretary of Planning, Information and Communications Technologies (ICT) Secretariat,
Modernization Government Secretariat, Argentina

4G-deployed infrastructures generate metadata provided by phone registrations systems. This data
provides the possibility to predict the behavior of users and the ways that devices are used throughout
the day. The definition of a metadata model and the design of artificial intelligence tools with learning
has the ability to exploit data and to provide a sample of the type of traffic and demand of services
that the user will request in each location visited. With this model we can predict the type of demand
needed to support the design of 5G networks.
Keywords – 5G, artificial intelligence, learning, network

Why machine learning?
Machine learning (ML) is a method of data analysis that automates analytical model building. It is a
branch of artificial intelligence (AI) based on the idea that systems can learn from data, identify
patterns and make decisions with minimal human intervention.

The core of the issue is the determination of patterns to be identified from 4G network activity as a
source of data to determine the behavior of the network as a system and the users as components,
with their own will that could modify the needs of services based in their activities.

An extraordinary amount of data is obtained in the exchange of information produced between the
network and the user’s devices. This scenario gives us the source of different kinds of information
that we could combine to obtain conclusions.

Over the system analyzed we must consider multiple variables that we can measure and register from
the logs that perform the processes of user registration, calls, hand over, IP assignment, flow of data,
and others.

The correlation between the multiple variables we are managing in a single analysis cannot be done
without the use of computational aid.

We are looking for patterns that we do not know of, and we need to discover, so the process of
learning from this repository of data is a new challenge because at the beginning of the process we
have no idea what relationship we are searching for.

The analysis along the timeline reveals the changes in the system and gives us an image of the
activities developed on the network, based on the change of values of each variable.
In this scenario, we need the support of machine-learning algorithms to detect and identify the
patterns on the network subject to analysis.

What type of machine learning models will be used with applications?
The model of learning will change at each stage of the work; the first step will be the use of the
“Supervised” model to validate the calculus that could predict values in a previous model like
bandwidth occupation, channel capacity or radio propagation; in this case, the variables must have
been identified and labeled previously.

– xv –