Since the inception of parent grant (July 2007), pattern recognition approaches such as neural networks, decision trees, Bayesian classifiers, and hidden Markov models have emerged as a viable and potentially more accurate alternative to traditional regression based cut-point methods. Whereas cut-point methods have proven to be modestly useful for broadly classifying the absolute intensity of physical activity, preliminary findings from the parent grant suggest that regression-based cut-point methods lack the accuracy required to provide valid point estimates of total and physical activity energy expenditure in field-based studies. Recently, pattern recognition approaches have been shown to be feasible and accurate using count data collected in 1-second increments by the widely used Actigraph accelerometer. The overarching goal of this competitive revision is to apply novel pattern recognition or machine learning data processing techniques to estimate physical activity type and intensity from high resolution accelerometer count data collected in children and adolescents as part of the ongoing parent grant. The proposed new aims are as follows: 1) To formulate and apply pattern recognition data processing techniques (multivariate hidden Markov models) to predict physical activity type and intensity from 1-second accelerometer count data collected in an age-diverse cohort children and adolescent;and 2) to compare the accuracy of physical activity intensity predictions provided by pattern recognition data processing techniques and traditional regression-based cut-point methods at baseline, and after 12-, 24-, and 36-months of follow-up.
The results of the proposed study will enable researchers to more effectively reduce and interpret accelerometer data collected in longitudinal studies of youth as well as population- based studies involving age diverse samples of children and adolescents. Moreover, the new aims will inform the development of new computational approaches for predicting physical activity energy expenditure in free-living children and adolescents. This development will be particularly important for observational and experimental studies wishing to investigate the contributions of physical activity and sedentary behavior in the development and maintenance of childhood obesity.
|Gammon, Catherine; Pfeiffer, Karin A; Pivarnik, James M et al. (2016) Age-Related Differences in OMNI-RPE Scale Validity in Youth: A Longitudinal Analysis. Med Sci Sports Exerc 48:1590-4|
|Montoye, Alexander H; Pfeiffer, Karin A; Suton, Darijan et al. (2014) Evaluating the Responsiveness of Accelerometry to Detect Change in Physical Activity. Meas Phys Educ Exerc Sci 18:273-285|
|Trost, Stewart G; Zheng, Yonglei; Wong, Weng-Keen (2014) Machine learning for activity recognition: hip versus wrist data. Physiol Meas 35:2183-9|
|Bassett Jr, David R; Rowlands, Alex; Trost, Stewart G (2012) Calibration and validation of wearable monitors. Med Sci Sports Exerc 44:S32-8|
|Robusto, Kristi M; Trost, Stewart G (2012) Comparison of three generations of ActiGraphýýý activity monitors in children and adolescents. J Sports Sci 30:1429-35|
|Trost, Stewart G; Wong, Weng-Keen; Pfeiffer, Karen A et al. (2012) Artificial neural networks to predict activity type and energy expenditure in youth. Med Sci Sports Exerc 44:1801-9|