Learning the Language of Human Activity in the Wild

• Activity Number: 161
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Korean International Statistical Society
• Author(s): Jiawei Bai*+ and Vadim Zipunnikov and Ciprian Crainiceanu
• Companies: Johns Hopkins University and Johns Hopkins University and Johns Hopkins University
• Keywords: accelerometer ; accelerometry ; physical activity ; movelet ; time series

Abstract

Predicting the type of activity performed by humans using accelerometry data is crucial in many scientific areas. Supervised statistical learning methods have shown great promise at predicting movement type when activity is observed in tightly controlled environments. However, in the wild (a.k.a. free-living) activity type has been very hard to predict, which dramatically limits the ability of researchers to describe the sphere of activity. We proposed to split very long accelerometry time-series data into short intervals, which are then clustered to identify the major components of activity and estimate movement complexity. Clusters obtained from the activity movelets in the wild were then compared and further quantified using the manually labeled data from the lab. This allows us to quantify movement using a fast, easy to implement, and highly interpretable activity prediction method, which allows to compare the physical activity structure and complexity of subjects.