Prediction of Activity Mode with Global Positioning System and Accelerometer Data

• Published in: Medicine and science in sports and exercise 2008-05, Vol.40 (5), p.972-978
• Main Authors: TROPED, Philip J, OLIVEIRA, Marcelo S, MATTHEWS, Charles E, CROMLEY, Ellen K, MELLY, Steven J, CRAIG, Bruce A
• Format: Article
• Language: English
• Subjects: Acceleration; Adult; Biological and medical sciences; Discriminant Analysis; Exercise - physiology; Female; Fundamental and applied biological sciences. Psychology; Geographic Information Systems - instrumentation; Humans; Male; Middle Aged; Pilot Projects; Predictive Value of Tests; Space life sciences; Vertebrates: body movement. Posture. Locomotion. Flight. Swimming. Physical exercise. Rest. Sports
• ISSN: 0195-9131; 1530-0315
• DOI: 10.1249/MSS.0b013e318164c407

The primary aim of this pilot study was to assess how well the combination of global positioning system (GPS) and accelerometer data predicted different activity modes. Ten adults (seven male, three female; 23-51 yr) simultaneously wore a GPS unit and accelerometer during bouts of walking, jogging/running, bicycling, inline skating, or driving an automobile. Discriminant function analysis was used to identify a parsimonious combination of variables derived from accelerometer counts and steps and GPS speed that best classified mode. A total of 29 bouts were used to develop this classification criterion. This criterion was validated using two datasets generated from the complete collection of minute-by-minute values from all bouts. Model development with "calibration" data showed that two accelerometer variables alone (median counts and steps) resulted in 26 of 29 bouts (90%) being correctly classified. Prediction of activity mode using counts and steps in a minute-by-minute "validation" dataset (N = 200) was 86.5%. Using three variables from the accelerometer and GPS (median counts, steps and speed) resulted in correct classification in 27 of 29 activity bouts in the "calibration" data (93%). In the "validation" dataset comprising 200 min, the combination of accelerometer counts and steps and GPS speed were able to correctly classify 91% of the observations. Walking and bicycling minutes were correctly classified most frequently (96%). In another "validation" dataset consisting of activity bouts, this combination of variables resulted in correct classification in 42 of 43 bouts (98%). This pilot study provides evidence that the addition of GPS to accelerometer monitoring improves physical activity mode classification to a small degree. Larger studies among free-living individuals and with an expanded range of activities are needed to replicate the current findings and further determine the merits of using GPS with accelerometers for mode identification.