QoE Assessment Model for Addis Ababa LTE Video Streaming Service Using Machine Learning Techniques
No Thumbnail Available
Date
2020-02
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
In today’s connected world, the availability of fast Internet access and penetration of
smart-phones has created an opportunity for the emerging of new telecom services.
Similarly, in Ethiopia, the improvement of technology brought a change from the
traditional services to more advanced communication like video streaming service. To
ensure whether the customers have satisfied for a given service or not, capturing user
Quality of Experience (QoE) is important. Traditionally, Internet Service Providers
(ISP)s monitor the network performance by collecting network key performance indicators
without involving users’ perception. However, user-perceived QoE estimation
is multidimensional, which is affected by different influencing factors. So, estimating
user-centric QoE based on Network-level QoS (NQoS) remains challenging tasks for
ISPs. Yet, QoE assessment model for video streaming services that map Quality of
Service (QoS) to QoE concerning users’ perception has not been performed in Ethiopia.
This thesis proposes video streaming QoE assessment models using machine learning
techniques to estimate user-perceived experience in the Long-Term Evolution (LTE)
network. The model predicts perceived QoE in a Mean Opinion Score (MOS), by
evaluating NQoS, Application-level QoS (AQoS) and contextually formulated survey
questionnaire. The models take NQoS metrics such as upload bit rate and download bit
rate in Megabits Per Second (Mb/s), latency and jitter in milliseconds (ms), and packet
loss in percentage. Content-type and resolution also considered from the application
level. Contrary to existing models for QoE prediction, the proposed model gives a good
estimation of the perceived quality with a minimum Mean Squared Error (MSE) of
7.74%; and Pearson and Spearman correlations of 97.94% and 97.43%, corresponding
to the measured QoE. The result obtained from the model shows that the average
MOS value is 2.79, which is below the recommended one. Accordingly, the proposed
model allows ISP to monitor the perceived QoE level accurately.
Description
Keywords
LTE, QoE, QoS, Video streaming, Machine Learning, Multivariate Linear Regression, Support Vector Regression