IEEE Computational Intelligence Magazine - May 2018 - 55

In this regard, we found several works in
the literature regarding the benefits of
big data in 5G networks, such as [12],
[13], where the authors identify different
sources of data that can act as an input to
the SON entity, e.g., to perform load
balancing and prediction operations,
among others.
In this work, we take advantage of
the MDT functionality (enhanced in
Release 11) by collecting measurements
indicating throughput and connectivity
issues to estimate network performance.
The exploitation of these huge amounts
of data is analyzed with regression analysis, where the primary goal is to predict
UE performance. We consider the analysis of our previous works [14]-[16],
where the exploitation of the huge
amount of data is analyzed with regression models to make better decisions for
management purposes. In particular, our
work in [15] focuses on two families of
regression models, linear and nonlinear.
A comparison was performed among
different models, selected on a basis of
low complexity and high accuracy. Prediction results were analyzed for different
kinds and amounts of UE measurements.
Based on the outcomes of that research,
in this new work, we focus on bagging
in combination with the support vector
machine (SVM), referred to hereafter as
the Bagged-SVM method.

Since we aim to find a trade-off between different goals, once the predictive
model has been built, we solve the SON
conflict by means of multi-objective
optimization, and in particular, multiobjective evolutionary algorithms that
use non-dominated sorting [5]. These
kinds of algorithms are based on metaheuristics that simulate the process of natural evolution and operate efficiently
when a large number of parameters need
to be configured simultaneously. While
the multi-objective problem formulation
presented herein can be solved by means
of RL [17], it is important to note that
following this approach, the agent needs a
good strategy to explore the environment
within a reasonable amount of time.
Therefore, in this paper we have supplied
the agent with NSGA-II, which is able
to maintain a spread of solutions with a
lower computational complexity.
Given this, the contributions of this
paper can be summarized as follows:
❏ We present a tool for an efficient
SON conflict resolution based on
network performance predictions. A
prediction model is created by extracting relevant information from the
mobile network.This allows the operator to anticipate system behavior to
make effective decisions in real-time.
Scalability is guaranteed since the processing complexity is shifted to the

creation stage of the model, allowing
fast performance evaluation when
optimization is running.
❏ In order to evaluate the proposed
scheme, we analyze, without loss of
generality, the MLB-MRO SON
conflict. The simulation results show
that the proposed scheme can not
only solve the SON conflict but also
improve overall system throughput.
III. General Self-Coordination
Framework

We consider a wireless network composed of a set M = {1, f, M} of
M = M cells regularly deployed with
inter-site distance D. On each cell
j = (1, f, N ), we consider N SON
functions running in parallel.We denote by
c ( j ) = (c (1j ), f, c (Nj)) the configuration
parameter vector of a cell j, with c (i j )
denoting the value for the parameter of
the SON function i, e.g., the transmission power (TXP), the antenna tilt
(TILT), the action to switch ON or OFF
the cell, handover parameters such as the
Cell Individual Offset (CIO), hysteresis
margin (HYS) or time to trigger (TTT).
The N SON functions are implemented
on every cell, which must send a request
to the SON coordinator and get a positive response in order to adjust some of
its parameters. These requests and
responses are represented in Figure 2

Self-Coordination Framework
Action Requests
Response

Cell1

SON1
SON2
SONN

Celli

Itf-N
SON1
SON2
SONN

c1i
i

c2
i

CellM

SON1
SON2
SONN

c1M
M

Figure 2 Each SON function requests the adjustment of different network parameters of cell i, and each one sends the action requests to the
self-coordination framework.

may 2018 | IEEE ComputatIonal IntEllIgEnCE magazInE

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - May 2018