Summary: In understanding travel demand, research often relies on global positioning system (GPS) data, which is becoming ubiquitous. The spatiotemporal nature of the GPS data allows for studying the behaviour at rich and detailed levels. Some of the steps in the data processing frequently rely on unsupervised machine learning techniques such as activity clustering and complex network analysis. Given the scale of the research phenomena in terms of the data volumes and geospatial extent, it is surprising that current processes still require much human intervention. The lack of automation and large scale data processing remains a challenge to routine decision support and more flexible research application. The work reported here contributes towards both these aspects. Through the use of a configurable, large-scale evolutionary algorithm, applying an expert-provided gold standard measure of expected clustering behaviour, a near-optimal clustering configuration is achieved while avoiding over-fitting. The optimisation workflow provides improved automation for repeatable and reproducible results. The workflow is even adaptable to clustering experiments with different optimisation configurations.