User adaptation in smart homes is always a challenging and long-term process. It becomes even more challenging when the user demand for facility from day one. Human activity recognition is one fundamental task of the adaptation process. There are many approaches, but currently the data-driven activity recognition approaches are currently the most promising way to address the sensor selection uncertainty problem in Smart Home research. However, one of the main weakness of the data-driven approach is data scarcity, also called the “cold start” problem. The very name “cold start” problem implies that a new inhabitant or user must wait for a certain time, perhaps an undesirably long time, to start benefitting from the new smart home’s services. This project’s aim is to design an integrated system that pre-determines user behaviour patterns using user-provided data. These patterns are used as part of a reasoning system to provide the functionality the user requires. Four methods are included to design the system: survey, transfer learning, activity recognition, and transfer learning.
