Predicting Human Movement with Multiple Accelerometers Using Movelets

Abstract

Purpose The study aims were: 1) to develop transparent algorithms that use short segments of training data for predicting activity types; and 2) to compare prediction performance of proposed algorithms using single accelerometers and multiple accelerometers. Methods Sixteen participants (age, 80.6 yr (4.8 yr); BMI, 26.1 kg·m−2 (2.5 kg·m−2)) performed fifteen life-style activities in the laboratory, each wearing three accelerometers at the right hip, left and right wrists. Triaxial accelerometry data were collected at 80 Hz using Actigraph GT3X+. Prediction algorithms were developed, which, instead of extracting features, build activity specific dictionaries composed of short signal segments called movelets. Three alternative approaches were proposed to integrate the information from the multiple accelerometers. Results With at most several seconds of training data per activity, the prediction accuracy at the second-level temporal resolution was very high for lying, standing, normal/fast walking, and standing up from a chair (the median prediction accuracy ranged from 88.2% to 99.9% based on the single-accelerometer movelet approach). For these activities wrist-worn accelerometers performed almost as well as hip-worn accelerometers (the median difference in accuracy between wrist and hip ranged from −2.7% to 5.8%). Modest improvements in prediction accuracy were achieved by integrating information from multiple accelerometers. Discussion and conclusions It is possible to achieve high prediction accuracy at the secondlevel temporal resolution with very limited training data. To increase prediction accuracy from the simultaneous use of multiple accelerometers, a careful selection of integrative approaches is required.

Objective Assessment of Physical Activity

“…In addition, the thorough training and quality control procedures should have minimized coding errors. The sample sizes of the study cohorts were modest, although comparable to most studies of this sort (11,13,17,25,30,33). The total minutes and days of data collected, however, were much greater than previous studies.…”

“…Our algorithms performed similarly to algorithms developed using other machine learning or statistical techniques (11,13,17,25,30,33). Most previous studies have been conducted in laboratory settings (13), or with some observed data in outdoor locations (30).…”

“…Most previous studies have been conducted in laboratory settings (13), or with some observed data in outdoor locations (30). No previous study has employed automatically captured observations over multiple days in free living or included driving and cycling.…”

Assessment of physical function and participation in chronic pain clinical trials: IMMPACT/OMERACT recommendations

“…28,201 Validity studies show that accelerometer counts are moderately to strongly associated with oxygen consumption during activity using direct and indirect calorimetry. 54,62,85 Further, actigraphy is strongly associated with measures of physical performance, such as the 6MWT, and predicts physical disability 82,109 and mortality. 15 Accordingly, actigraphy provides valuable objective data on the intensity and extent of physical activity and is increasingly used in studies of chronic pain.…”

Hip and Wrist Accelerometer Algorithms for Free-Living Behavior Classification

“…An advantage of wrist accelerometers is that they may improve compliance, and can be worn 24 hours without removal for sleep (33). Previous studies have shown that a wrist accelerometer can be used to classify PA behaviors (12, 19) and estimate overall PA (10) and intensity thresholds (13, 23, 25). More research is needed, however, to evaluate the validity of wrist-worn devices for measuring PA in free-living.…”