The use of accelerometers for the measurement of physical activity in youth has increased substantially over the last decade. Accelerometers are typically worn on the waist, however recently alternative placement sites such as the wrist have become more popular. For example, the National Health and Nutrition Examination Survey (NHANES) is currently using the wrist location for their accelerometer physical activity measurements. Currently there is a lack of information about best approaches for analyzing accelerometer data from the wrist location with the ActiGraph accelerometer. In addition, advances in accelerometer technology allow for utiliza-tion of the raw acceleration signal with machine learning algorithms, which have the capability to predict activi-ties and improve the estimates of energy expenditure over previously used methods. Members of this research group have shown improved activity classification and energy expenditure prediction in youth using Bipart, ver-sus commonly used approaches such as Artificial Neural Networks and Support Vector Machine. However, this preliminary work was based on accelerometer count data and structured lab activities with only a hip worn ac-celerometer. In addition, there are limitations to applying lab based models to true free-living activity. For ex-ample, free-living activity is not performed in structured bouts rather it is performed in micro-bouts over the course of the day which is problematic for methods designed to look at a string of data over a specified time period. There is minimal work currently on how to first segment bouts of free-living activity before classifying activity type and predicting energy expenditure. This proposal will extend our previous work using more ad-vanced methods to analyze raw acceleration data (80 Hz) using a single wrist or hip worn accelerometer. One hundred youth will be measured during a semi-structured simulated free-living period (development group) and 200 youth will be measured during true free-living activity during an after-school program and at home (valida-tion group). Measurements will include indirect calorimetry for energy expenditure and direct observation for activity type. The specific aims of the study are to: 1) develop and validate machine learning algorithms to segment bouts of activity during free-living activity in youth using: A) a hip worn accelerometer or B) a wrist worn accelerometer, and 2) develop and validate machine learning algorithms to classify physical activity type and estimate energy expenditure in youth, during free-living activity using: A) a hip worn accelerometer or B) a wrist worn accelerometer. Results from these studies will have direct and immediate impact for physical activity researchers utilizing accelerometers as well as analysis of NHANES wrist accelerometer data by providing ac-curate and precise methods for bout segmentation, prediction of activity type and energy expenditure.
Activity recognition and intensity estimation in youth from accelerometer data aided by machine learning
Xiang Ren*, Wei Ding, Scott E. Crouter, Yang Mu, Rong Xie, Applied Intelligence, 45(2): 512-529, 2016.
Bipart: Learning Block Structure for Activity Detection
Y. Mu*, H. Lo*, W. Ding, K. Amaral**, S. E. Crouter, IEEE Transactions on Knowledge and Data Engineering (TKDE), Pages 2397-2409, Volume 25, Issue 10, doi: 10.1109/TKDE.2014.2300480, Oct. 2014.
Local Discriminative Distance Metrics and Their Real World Applications
Y. Mu*, W. Ding, Ph.D. Forum in conjunction with IEEE International Conference on Data Mining (ICDM), Dallas, TX, USA, December, 2013.
Discriminative Accelerometer Patterns in Children Physical Activities
Two-Tiered Machine Learning Model for Estimating Energy Expenditure in Children
Wei Ding, Associate Professor, Computer Science Department, UMass Boston
Ping Chen, Associate Professor, Computer Science Department, UMass Boston
Scott Crouter, Associate Professor, College of Education, Health, and Human Sciences, University of Tennessee.
Kevin Amaral, PhD student, Computer Science Department, UMass Boston
Matthew Almeida, PhD student, Computer Science Department, UMass Boston
Zihan Li, PhD student, Computer Science Department, UMass Boston
Dr. Yang Mu, graduated, Facebook, Inc.
Dr. Henry Lo, graduated, McKinsey & Company.