Smart Home Adaptation Based on Explicit and Implicit User Feedback By

by Parisa Rashidi, M.S. Washington State University December 2007 Chair: Diane J. Cook In current work we introduce CASAS, an adaptive smart home system that utilizes machine learning techniques in order to dynamically adapt to user advice or changes in daily routine activities. The main components of CASAS include a frequent and periodic activity miner (FPAM), a hierarchal activity model (HAM), a dynamic adapter and CASAS’s user interface and visualizer, CASA-U. The FPAM algorithm discovers arbitrary length periodic and frequent patterns from the resident’s daily activities efficiently by utilizing the minimum description length principle. HAM, as a hybrid model of a decision tree combined with a Markov decision process, provides a hierarchal abstraction of patterns while utilizing temporal information such as temporal relations, temporal granules, start time and duration distribution. HAM is used to identify potential automations. The dynamic adapter component allows HAM to dynamically adapt to the user’s explicit feedback (advice) or implicit feedback (changes in daily routine activities) based on four techniques of explicit manipulation, explicit rating, explicit request and smart detection. It exploits guidance-based learning and observation-based learning along with the Activity Adaptation Miner (AAM) to adapt to these types of feedback. Finally, to provide users with greater control over their personal environment and to provide a framework for explicit manipulation and rating of suggested automation policies, a user interface is provided that enables residents to navigate through a map of the home, view a history of events, modify the events and provide guidance to the smart home’s automation policies. Integrating all these components together, the architecture of CASAS is provided that shows how resident interactions in a smart home can be automated and continually adapted to explicit or implicit changes in the resident’s patterns. We also show the results of our successful experiments with CASAS on both synthetic and real world data, besides a usability test of CASA-U.