Optimising an agent-based model to explore the behaviour of simulated burglars