A Tutorial on Data-Driven Quality of Experience Modeling With Explainable Artificial Intelligence

Data-driven Quality of Experience (QoE) modeling using Machine Learning (ML) is a key enabler for future communication networks as it allows accelerated and unbiased QoE modeling while autonomously adapting to changes over time. Traditional ML techniques only provide black-box QoE models, meaning that the models internal reasoning is hard to understand. This makes it difficult not only to gain valuable insights into the data and the underlying QoE influence factors, but also to build trust in the predictions of the QoE model. To address these challenges and enhance performance, we identify four critical aspects for any data-driven QoE model: explainability, context adaptability, quantification of uncertainty, and data decentralization. In this work, we focus on eXplainable Artificial Intelligence (XAI), which is a promising solution to obtain explainable models allowing to address all critical aspects for data-driven QoE modeling. This tutorial introduces and explains relevant concepts from XAI and related fields. We apply these techniques to two realistic use cases, video streaming QoE modeling and web QoE modeling, for two different modeling tasks, classification and regression, and discuss their value in the context of XAI-based data-driven QoE modeling. In addition, to improve the readers understanding, to ease the application of presented techniques for own use cases, and to foster research in the field of data-driven QoE modeling with XAI, we make Python Jupyter Notebooks available.