Novel Technique for Assessing Physical Activity Using Accelerometer Data

1394 PURPOSE: Objective assessment of physical activity in a field setting is a crucial component of many epidemiological studies. The use of accelerometers for this purpose is widespread, but the utility of this method is frequently hampered by misclassification of activity levels based on the data from accelerometers. Estimates of activity type, rather than intensity, could eventually provide a better overall picture of physical activity related energy expenditure. We hypothesized that a stochastic model would have the capability to distinguish different activity types using accelerometer data. Therefore, the purpose of this investigation was to develop an algorithm to estimate type of activity using accelerometer data. METHODS: To accomplish this task, a Hidden Markov Model (HMM) was created using training data from a uniaxial accelerometer worn by a single subject who participated in walking, running, sitting quietly, and vacuuming for 10 minutes each. HMMs are simple stochastic time series models for discrete data. This model was then tested against data from a uniaxial accelerometer worn by a different subject performing the same activities. RESULTS: The HMM correctly identified 100% of walking and sitting epochs, 99.7% of running epochs, and 83.7% of vacuuming epochs. In all cases, the misclassified epochs were estimated to be walking. CONCLUSIONS: This approach shows substantial promise as a technique for assessing physical activity monitor data. The novel approach of estimating activity mode, rather than activity level, may allow for more accurate assessment of physical activity in the field, and provide valuable information regarding the effects of intervention programs.