Modeling subjective tests of quality of experience with a Generalized Linear Model

Abstract

Subjective tests of telecommunication services engaging a panel of subjects aim to analyze how service users experience their quality (QoE) with varying parameters of both a network and/or application itself. Subjects express their opinion on a few-grade categorical scale during tests performed in a controlled environment. The proposed Generalized Linear Model (GLZ)1 relates probabilities of particular QoE levels to independent network and application explanatory variables in a statistically credible way. The GLZ model is executed in several steps including maximum likelihood estimation of model parameters, selection of the most effective model with the Bayesian information criterion, model validation with a chi-squared test, and last but not least confidence interval estimation. Monitoring a QoE level with a GLZ model provides to operators an effective tool towards a mid- and long-term adaptation of both network and/or application parameters in order to follow users' experience and expectations.