A Measurement-based Quality of Experience Model for Mobile Video Streaming: A Hidden Markov Model Approach
| dc.contributor.advisor | Dereje Hailemariam | |
| dc.contributor.author | Amel Salem Omer | |
| dc.date.accessioned | 2026-04-04T11:36:31Z | |
| dc.date.available | 2026-04-04T11:36:31Z | |
| dc.date.issued | 2025-12 | |
| dc.description.abstract | Demand for mobile multimedia services is rising; as a result, operators routinely monitor network performance using Quality of Service (QoS) metrics such as latency, throughput, and packet loss. While these metrics indicate network performance, they do not fully reflect end-user experience, particularly under fluctuating signal strength, bandwidth constraints, and diverse usage conditions. Though difficult to measure, Quality of Experience (QoE) captures users perceived service quality, which determines their satisfaction and loyalty. The gap between measurable QoS and subjective QoE remains a key challenge, and bridging it is vital for improving service delivery. This dissertation addresses the QoS–QoE gap using a data-driven, machine-learning approach based on data collected from a Mobile Network Operator (MNO). For mobile video streaming, it proposes a Hidden Markov Model (HMM)-based model to predict user-perceived QoE from measurable QoS parameters. A custom Android-based data collection mobile app, called iNET, was developed to collect synchronized user-side network metrics and subjective feedback from 550 users across different devices, usage times, and locations in Addis Ababa. The data are used to build a QoS to QoE mapping model. The modeling builds on our earlier experience in analyzing network accessibility, retainability, and congestion using Markov Chain and HMM based models trained on real-world datasets from an operator’s Network management system (NMS). This experience was vital for the proposed QoS-to-QoE mapping, which is the core of this dissertation. Central to QoS–QoE mapping is identifying the hidden and observable state in the HMM. QoE is typically latent from an MNO perspective, which aligns with treating it as the hidden state in HMM terminology, while network-side QoS metrics as observations. However, in the real system, QoS conditions influence QoE, which can motivate modeling QoS as the underlying hidden state and QoE as the outcome. This creates a modeling paradox between adhering to HMM emission terminology and reflecting the causal behavior of the QoS–QoE process, and it similarly affects whether QoS should be treated as hidden or observable. Accordingly, we evaluated both formulations by alternately treating QoE and QoS as hidden states and comparing their empirical performance. As the HMM requires a single sequence, we applied Principal Component Analysis (PCA) to reduce the dimensionality of QoS features and then used K-means to quantize them into a single cluster-label sequence. The HMM-based model achieved 98.04% accuracy, outperforming baseline Random Forest (96.15%) and Support Vector Machines (SVM) (49.03%) models. The results show that HMM-based modeling enables accurate QoE prediction and highlight the potential of the iNET measurement tool for use by MNOs. | |
| dc.identifier.uri | https://etd.aau.edu.et/handle/123456789/8054 | |
| dc.language.iso | en | |
| dc.publisher | Addis Ababa University | |
| dc.subject | iNET | |
| dc.subject | QoS | |
| dc.subject | QoE | |
| dc.subject | HMM | |
| dc.subject | Markov Chain | |
| dc.subject | Long Term Evolution (LTE) | |
| dc.subject | Mobile Network | |
| dc.subject | Modeling | |
| dc.subject | Prediction | |
| dc.title | A Measurement-based Quality of Experience Model for Mobile Video Streaming: A Hidden Markov Model Approach | |
| dc.type | Thesis |