Figure 6.
Classification performance for full and reduced sensor counts. Performance was computed using LDA and data from with progressively reduced sensor counts. Seven sensors (pelvis, sternum, head, and the active shoulder, upper arm, forearm, and hand) gave the best classification performance, with a performance drop-off at more or fewer sensors. IMU data consistently supported higher classification than accelerometery data, achieving PPV 92.5 vs. 82% at the seven sensors.