Estimating QoE from Encrypted Video Conferencing Traffic.

Area of Science:

• Computer Science
• Network Engineering
• Information Security

Background:

• Internet security relies on traffic encryption, which hinders analysis of applications like video delivery optimization and Quality of Experience (QoE) estimation.
• Existing QoE prediction models primarily focus on clear-text traffic and often provide low-resolution, categorical predictions, leaving the video conferencing (VC) domain underexplored.

Purpose of the Study:

• To address the challenges in QoE assessment for encrypted traffic in video conferencing (VC) applications.
• To develop precise, continuous QoE predictions for VC, moving beyond broad categories.
• To analyze encrypted traffic data for improved QoE estimation.

Main Methods:

• Analysis of a large dataset comprising Zoom sessions.
• Training and evaluation of five classical machine learning (ML) models.
• Development and training of two custom deep neural networks (DNNs).
• Prediction of three key QoE indicators: frames per second (FPS), resolution (R), and Naturalness Image Quality Evaluator (NIQE).
• Utilizing a 10-fold cross-validation technique for robust model assessment.

Main Results:

• Achieved mean error rates of 8.27% for FPS prediction.
• Achieved mean error rates of 7.56% for resolution (R) prediction.
• Achieved mean error rates of 2.08% for Naturalness Image Quality Evaluator (NIQE) prediction.

Conclusions:

• The developed models demonstrate significant advancements in QoE assessment for encrypted traffic within VC applications.
• Precise, continuous QoE predictions are achievable even with encrypted video conferencing data.
• This research provides a foundation for enhanced network management and user experience in video conferencing services.