Identifying the underlying dynamics of chemical process systems from experimental data is complicated, owing to a mixture of influences that cause erratic fluctuations in the time series. These influences can be notoriously difficult to disentangle. The development of process models is usually subject to considerable human judgment and can therefore be very unreliable. This is especially the case when the model priors are unknown and the model is validated empirically, such as with cross-validation or holdout methods. A case study shows that more reliable identification of systems is possible by using surrogate methods to classify the data, as well as to validate models derived from these data.