| Mobile phones are ubiquitous and increasingly capable, with sophisticated sensors, network access, significant storage and processing power and access to a wide range of application data. They can improve the range and quality of their services by acquiring and using models of their context, including the activities in which their users are engaged. This thesis explores the use of supervised machine learning techniques for predicting a Smartphone user's activities from available sensor data. We have specifically concentrated on applying classifiers and ensembles using hidden markov models for activity recognition. Our classifiers predict a user's current activity from among a set of conceptual activity classes such as sleeping, traveling, playing, working, and chatting/watching TV. We have experimented with and evaluated the effectiveness of different approaches on data collected on Android Smartphone by university faculty and students. |
