For service providers and operators, successful root cause analysis is essential for satisfactory service provisioning. However, reasons for sudden trend changes of the instantaneous Quality of Experience (QoE) may not always be immediately visible from underlying service- or network-level monitoring data. Thus, there is the challenge to pinpoint such moments of change in provisioning, and model the impact on instantaneous QoE, as a lead in root cause analysis. This work investigates the potential of Machine Learning (ML) of deriving time-interval-based models for instantaneous QoE ratings, obtained from a set of publicly available rating traces. In particular, the paper demonstrates the capability of the ML algorithm M5P to model trends of instantaneous QoE through model trees, consisting of piecewise linear functions over time. It is shown how and to which extent these functions can be used to estimate moments of change. Furthermore, the model trees support earlier assumptions about exponential shapes of instantaneous QoE over time as reactions to sudden changes of provisioning, such as video freezes. © 2019 IEEE.