This classic paper was a tour de force demonstration of a practically deployable activity recognition system. Previous work had focused on heavy instrumentation in the lab; this was on essentially commodity phones, and over a long period of time. It essentially paved the way for FitBit and Apple Health monitoring. In addition, it widened the design space of behavior design interfaces encouraging users to change their behaviors (e.g., walk more, eat less, sleep more)
Tanzeem Choudhury's work is at the forefront today of activity recognition systems. This recent paper is interesting because it tackles a problem of detecting an issue — mental health changes — before it advances. It demonstrates how far the field has come: it uses more rigorous, modern evaluation techniques, and connects more directly with the clinical literature.