Support Vector Regression for METs of Exergaming Actions

Mortazavi, Bobak ;   Pourhomayoun, Mohammad ;   Alshurafa, Nabil ;   Chronley, Michael ;   Lee, Sunghoon Ivan ;   Roberts, Christian K ;   Sarrafzadeh, Majid

Sedentary behavior is a root cause of several chronic conditions affecting health of adults and children in the United States and worldwide. The chronic conditions that result from this cause not only health concerns for these individuals but significant economic burden. Exergaming, or the merger of exercise and health information with video games, presents a solution that attempts to address the sedentary behavior of adults and children by making physically interactive video games that increase energy expenditure. Such games, particularly those that use the body as the controlling device for the game through the use of accelerometers, have elicit moder- ate levels of physical activity when measuring the metabolic equivalent of task (MET) of the associated activities. This work presents the support vector regression scheme in order to better correlate accelerometer measurements with MET values. Energy expenditure data collected on 14 individuals and their accelerometer data have regressions with the mean absolute difference (error) of the associated MET approximations is under 2 and as low as 0.58 for full gameplay, an improvement of well over 1 MET for all activities over related work.