This paper describes Actitracker, a smartphone-based activitymonitoring service that helps people ensure that they receive sufficient activity to maintain proper health. Unlike most other such services, Actitracker requires only a smartphone (no watches, bands, or clip-on devices). This free service allows people to set personal activity goals and monitor their progress toward these  goals. Actitracker uses data mining to generate its activity recognition models. It initially uses a universal/impersonal model that is generated from labeled activity data from a panel of users, but will automatically generate, and deploy, much more accurate personalized models once a user completes a simple training phase. Detailed activity reports and statistics are maintained on the Actitracker server and are available to the user via a secure web interface. Actitracker has been deployed for several months and currently has over 250 registered users. This paper discusses user experiences with the service, as well as challenges and tradeoffs associated with building and deploying the service

Jeffrey W. Lockhart, Gary M. Weiss, “The Benefits of Personalized Models for Smartphone-Based Activity Recognition,” Proceedings of the SIAM International Conference on Data Mining, Philadelphia, PA (2014).