Enrichment of Machine Learning Based Activity Classification in Smart Homes Using Ensemble Learning

Abstract

Data streams from various Internet-Of-Things (IOT) enabled sensors in smart homes provide an opportunity to develop predictive models to offer actionable insights in form of preventive care to its residence. This becomes particularly relevant for Aging-In-Place (AIP) solutions for the care of the elderly. Over the last decade, diverse stakeholders from practice, industry, education, research, and professional organizations have collaborated to furnish homes with a variety of IOT enabled sensors to record daily activities of individuals. Machine Learning on such streams allows for detection of patterns and prediction of activities which enables preventive care. Behavior patterns that lead to preventive care constitute a series of activities. Accurate labeling of activities is an extremely time-consuming process and the resulting labels are often noisy and error prone. In this paper, we analyze the classification accuracy of various activities within a home using machine learning models. We present that the use of an ensemble model that combines multiple learning models allows to obtain better classification of activities than any of the constituent learning algorithms.