Modeling Transition and Mobility Patterns

Abstract

We present a solution to model user transitions and mobility patterns without the need of accessing any cloud services, and thus completely preserving user privacy. Our algorithm relies solely on the sensor inputs of the mobile device to gather environmental fingerprints. A real-time hierarchical clustering algorithm efficiently organizes the individual signatures into a hierarchy of meaningful significant locations at various time scale. By applying (normalized) information measure and neural network based learning, we are able to identify the most salient transition patterns that best characterize the mobility data of the user. Our algorithms are completely online, and do not rely on any networked resources. Thus, user can gain insight to their own mobility activities, and has total control of how this information is to be shared with other applications. We will demonstrate using real-life data the effectiveness and efficiency of our approach. Several appealing visualizations will be showcased. The resulting transition models can be utilized towards better user experience for a variety of mobile applications such as activity scheduling and travel route planning.