Page 26 - Proceedings of the 2018 ITU Kaleidoscope

2018 ITU Kaleidoscope Academic Conference

The proposed methodology represents a major challenge
because of the unpredictable nature of the scenarios
considered, with networks sharing the spectrum (even
working in unlicensed bands) and a response totally
dependent on the behavior of users and many other
contextual and non-contextual agents not controlled by the
providers. Addressing this challenge through ML
constitutes the novelty of this proposal.

The remainder of the paper is organized as follows: Section
2 summarizes related work on QoS management
approaches in wireless scenarios considering the use of
ML-based models, and also standardization-related works
are reviewed. Section 3 describes the QoS management
model that is adopted and defines the QoS-QoE-QoBiz
relations to be considered. In Section 4, the proposed
methodology to implement the model in NGW scenarios
through ML mechanisms is described. Section 5 illustrates
the experiment carried out to validate the proposed
methodology and, finally, Section 6 contains some
conclusions and final remarks.

2. BACKGROUND
Figure 1. Framework for modeling the perceived QoE
Many studies and standardization efforts related to QoS and using a big-data analytics approach [9]
QoE have been conducted in recent years. Most of these
show that the assessment of QoS has moved away from The framework presented in [9] (Figure 1) is a step ahead
network performance (NP) in favor of QoE, related to the of the other works. In the proposed framework, “the
subjective perception of end users. However, as stated in process of estimating or predicting the perceived QoE
the Recommendation ITU-T G.1000 on the QoS framework based on the datasets obtained or gathered from the mobile
[8], the great challenge to success when deploying a QoS network to enable the mobile network operators effectively
management model is to embrace all the different to manage the network performance and provide the users
QoS-related aspects (e.g. NP, QoS, QoE, QoBiz) and, more a satisfactory mobile Internet QoE” is described. The
importantly, quantifying the relationships between them. state-of-the-art included in this work covers a large set of
This may become a difficult task when dealing with experiments in different scenarios and for different
next-generation wireless ecosystems, where many applications and services in relation to modeling QoE with
unpredictable instances may have an influence on the user´s ML algorithms. Their study about the dimensions of the
experience. QoE influence factors reviews a great number of scientific
proposals. After the analysis, the authors concluded that
In view of this, some recent studies suggest using big data three QoE dimensions (human, system, and context) should
analysis and ML algorithms for modeling the QoE-QoS be considered when modeling QoE (Figure 2).
relationship. ML techniques may be useful to infer rules
from big data analysis and identify the KPIs/KQIs that will Aroussi and Mellouk presented in [10] an interesting survey
lead to automatically estimating the quality as perceived by about ML-based QoE-QoS correlation models. They
users based on the QoE influence factors. Selecting the suggest that supervised or semi-supervised learning models
suitable learning algorithm may be critical to obtaining provide a better fit for the QoE-QoS correlation modeling
reliable results. than non-supervised ML.

2.1 QoE models and machine learning

There is limited research on the modeling of QoE with ML
for next-generation mobile scenarios. Nevertheless, there
are some recent studies focused on analyzing and using the
data captured from the network and the users' surveys to
model and to enhance the QoE [9-11].

Figure 2. QoE dimensions and influence factors [9]

– 10 –