Ubiquitous Computing Smartwatch Data

Project #2145 on iSENSEProject.org

To help prevent onset arthritis, users with a smartwatch can install wrkr (an app in progress for COMP.5800) to help them track when it is time to take a quick break from sitting at a computer and do a few wrist exercises, guided by a Leap Motion device at our website (again, in progress).

This preliminary data will be used to train a machine learning model that will help us detect when a user is sitting at a keyboard versus when he/she is not.  We hope, therefore, that a logistic regression model will suffice.  It is unlikely a clustering algorithm (e.g. K-means) is sufficient, and neural networks are non-trivial.  Accelerometer data collection is done on the watch that holds a partial wakelock as it transmits 10-second bursts of data over low-power Bluetooth to the mobile device that does the more intensive processing and calculations on the data.  The data sets you see below are thus roughly 10 seconds worth of data per data set (50 points per set recording at roughly 5Hz).