The Independent Lifestyle Assistant (ILSA) is agent-based system to aid elderly people to live longer in their homes. One of the biggest challenges for ILSA is that every installation will (1) be in a home with a different layout and suite of sensor and actuator capabilities, and (2) supporting a technophobic client with unique capabilities, needs and care-giving support network. ILSA therefore must be able to adapt to its environment over time. This paper describes our approach to modelling one particular aspect of the ILSA domain: using sequential pattern learning to learn models of human behaviour. We describe the problem, our enhancements to the basic algorithm, and present experimental results collected from four test sites~