Semi-Supervised Learning in Inferring Mobile Device Locations

With the development of mobility technology, location information has become collectible by various positioning technologies. Different positioning technologies have their advantages and limitations. In particular, location accuracy varies among different technologies. In this paper, we propose semi‐supervised learning in inferring low‐accuracy location data density from high‐accuracy location data density. We focus on the enormous amount of low‐accuracy cell tower triangulation (CTT) calculated mobile device location data and the small amount of high‐accuracy assisted global positioning system (AGPS) pinpointed location data. The CTT and AGPS mobile device location data are collected for each cell tower that serves the devices, and then, the actual distribution is learned from both CTT and AGPS data by semi‐supervised learning, and the likelihood for low‐accuracy CTT location can be used as an accuracy indicator. The proposed method takes the advantage of the existing extensively collected location data and augments it by a machine‐learning algorithm. The method does not only complement the downside of one technology with the other technology but also considers the computing efficiency by extracting features that can segment the large data naturally, which is the essential in divide‐and‐conquer distributed big data processing. This big data approach improves the location accuracy statistically without the added complexity and cost of upgrading or replacing mobile networks or devices. Moreover, also, the proposed method focuses on the location density alignment, which avoids tracking individual user devices and preserves user privacy.