The Impact of Walking and Resting on Wrist Motion for Automated Detection of Meals

Abstract

This article considers detecting eating in free-living humans by tracking wrist motion. We are specifically interested in the effect of secondary activities that people conduct while simultaneously eating, such as walking, watching television, or working. These secondary activities cause wrist motions that obfuscate those associated with eating, increasing the difficulty of detecting periods of eating. We collected a large dataset of 4,680 hours of wrist motion from 351 participants during free living. Participants reported secondary activities in 72% of meals. Analysis of wrist motion data revealed that the wrist was resting 12.8% of the time during self-reported meals compared to only 6.8% of the time in a cafeteria dataset, whereas walking motion was found 5.5% of the time during meals in free living compared to 0% in a cafeteria. Augmenting an eating detection classifier to include walking and resting detection improved accuracy from 74% to 77% on our free-living dataset ( t [353] = 7.86, p < 0.001). Although eating detection could be improved using more sophisticated machine learning methods or sensor modalities, all approaches would be affected by secondary activities, as they affect the labeling of data itself. Our work suggests that future work should collect detailed ground truth on secondary activities being conducted during eating, as these activities could hold insights into when an eating activity starts or stops in the absence of video-based ground truth.

Combining ecological momentary assessment, wrist-based eating detection, and dietary assessment to characterize dietary lapse: A multi-method study protocol

“…Our eating inference algorithms segment the IMU data at peaks of wrist motion, 29 and then these segments are classified as eating, walking, resting, or other, using a Bayesian classifier. 43 Classification of eating is further described in the Statistical Analysis section. Once eating episodes are inferred, we calculate duration and timing of eating.…”

“…Prior work, completed with other types of wrist-mounted devices, has developed and refined algorithms that can infer eating episodes and characteristics from wrist motion and wrist velocity data. 28 – 31 We are now extending these algorithms to infer eating from wrist data collected by the ActiGraph. Because the composition of foods consumed during lapses is an important and understudied facet of lapse behavior, participants also complete periodic 24-hour dietary recalls via structured telephone interview to measure the composition of all food and beverages consumed.…”

“…As such, our recent work has focused on developing and refining algorithms that can infer eating episodes from continuously collected wrist data (thus eliminating the need for a button press). [27][28][29][30][31] Our methods for eating inference and free-living bite counting have been developed and refined over five studies (each using differing wrist-based sensors, e.g., InertiaCube3 sensor, iPhone 4, Bite Counter device, STMicroelectronics MEMS sensors, Shimmer3 device). In the first, we counted bites taken during a meal with 86% sensitivity among 47 participants wearing a wrist device during an uncontrolled eating setting in which they brought their own meal to a lab.…”

“…As such, our recent work has focused on developing and refining algorithms that can infer eating episodes from continuously collected wrist data (thus eliminating the need for a button press). 27 – 31 …”

“… 27 Fifth, our most recent study evaluated 4,680 hours of data across 351 participants who wore a wrist device for 1 day during free-living and used an activity log to self-report ground truth of eating instances and secondary activities (i.e., walking and resting). 30 This work capitalized on accounting for secondary activities detected in the wrist data to achieve 77% weighted accuracy of inferring eating episodes. Further analysis showed that the improvements over our prior methods (which achieved 74% weighted accuracy in this sample) 29 due to accounting for walking and rest were significant (p < .001) and could be attributable to a 23% reduction in false negatives.…”