A Framework for User Routine Discovery in Smart Homes

Abstract

With recent advancements in ubiquitous sensor technologies and smart environments, routine discovery has emerged as a highly studied problem because of its applications in smart home automation, anomaly detection and assisted living for elderly. Routines can be categorized on the basis of some discernible properties like frequency, recurrence and periodicity. Commonly researched solutions to discover routines/activities of daily living are based on topic modeling or association rule mining (ARM). Both ARM and topic modeling based approaches are computationally expensive and do not acknowledge routine categorization, hence are not guaranteed to discover all different kinds of routines. In this paper we propose a simple, efficient and unsupervised framework for routine discovery in smart homes based by exploiting properties of routines which categorize them. We define and leverage such properties to discover routines utilizing an iterative algorithm, where routines from a specific category are discovered at each iteration. We have demonstrated a routine discovery system FIBI by implementing the framework and have achieved more than 80% recall and more than 90% precision across different categories of routines defined under FIBI, across three different market segments - social climber, affluent nester and urban dweller.