Designing and evaluating active learning methods for activity recognition
Salikh Bagaveyev, Diane J. Cook · 2014
Activity recognition in smart home environments is a crucial step towards fully autonomous assistance and health monitoring. Due to the high variance in house configurations and sensor placements, it is important to collect and label sample sensor data that will be used to train a learning algorithm. Ground-truth activity labels must therefore be provided in some manner for this sample data. The abundance of sensor data makes it infeasible to label all of the data, and active learning can be used to intelligently pick the most informative data to be labeled. In this paper, we describe several active learning methods that we designed and implemented for their applicability to the activity recognition task. We evaluate the methods using the CASAS smart home sensor data and present a crowd-sourcing application for annotation.