Next: Model Overview
Up: Cluster Weighted Modelling: A
Previous: Cluster Weighted Modelling: A
From September 22nd to 25th, 1997, seven soldiers at the Fort Benning army base were equipped with sensor packages which measured a number of biometric and situational parameters: GPS (position, elevation, heading and speed), heartrate, core body temperature, and step count. They engaged in a number of activities, including a 20km road march, rest periods, and a number of assaults on training targets.
These data were sampled at a rate of 1 sample every 5 seconds, continuously, using the ``Marathon Man'' rig developed by the Personal Information Architecture group at the MIT Media Lab.
Given this biometric data, it is desirable to be able to discern the state of the soldier, especially to monitor for anomolies that might warrant medical attention, and to be able to automatically evaluate what activity a soldier is engaged in. Given the sample data, the second problem was deemed more immediately approachable.
Examining the data available by hand made it clear that a resting/working discriminator would be a trivial linear discriminant based on speed, as most of the work activity involved marching, and the rest activity involved little movement. The goal was to therefore get a attacking/not-attacking discriminator.