School of Human Movement Studies, The University of Queensland
Machine Learning for Activity Recognition: Hip versus Wrist Data
- Presented on June 18, 2013
Introduction: Wrist-worn accelerometers are convenient to wear and are associated with greater compliance. However, validated algorithms for predicting activity type and/or energy expenditure from wrist-worn accelerometer data are lacking.
Purpose: To compare the activity recognition rates of an activity classifier trained on raw tri-axial acceleration signal (30 Hz) collected on the wrist versus the hip.
Methods: 52 children and adolescents (mean age 13.7 +/- 3.1 Y, 28 boys, 24 girls) completed 12 activity trials that were categorized into 7 activity classes: lying down, sitting, standing, walking, running, basketball, and dancing. During each trial, participants wore an ActiGraph GT3X+ tri-axial accelerometer on the right hip and the non-dominant wrist. For both hip and wrist data, features were extracted from 10-s windows and inputted into a regularized logistic regression model using R (Glmnet + L1). The average classification accuracy was calculated over 30 training-validation-testing iterations.
Results: Classification accuracy, averaged over all 7 activity classes, for the HIP and WRIST algorithms was 91.0 +/- 3.1% and 88.4 +/- 3.0%, respectively. The HIP model exhibited excellent classification accuracy for sitting (91.3%), standing (95.8%), walking (95.8%), and running (96.8%); acceptable classification accuracy for lying down (88.3%) and basketball (81.9%); and modest accuracy for dance (64.1%). The WRIST model exhibited excellent classification accuracy for sitting (93.0%), standing (91.7%), and walking (95.8%); acceptable classification accuracy for basketball (86.0%); and modest accuracy for running (78.8%), lying down (74.6%) and dance (69.4%).
Conclusion: Activity recognition was marginally higher using tri-axial acceleration signal from the hip versus the wrist. However, the small difference in performance may not be of practical significance in field-based studies. Both algorithms achieved acceptable classification accuracy.
Supported by: NIH R01 NICHD 55400