Food watch: detecting and characterizing eating episodes through feeding gestures

摘要: Obesity, caused primarily by overeating, is a preventable chronic disease yielding staggering healthcare costs. To detect overeating passively, machine learning framework was designed to and accurately count the number of feeding gestures during an eating episode characterize each with gesture count. With ubiquitous nature wrist-worn sensors, existing literature has focused on detecting eating-related episodes that are at least f ve minutes long. In this paper, our objective show potential commercial smartwatches be used in detection short durations confounded other activities daily living order truly capture all field The effect time-series segmentation sensing configuration accuracy characterizing then analyzed. Finally, effects personalized generalized models predicting compared. Results demonstrate large within-subject variability eating, where user-independentmodel yields 75.7% average F-measure, whereas user-dependent modelyields 85.7% F-measure. This shows clustering count, resulting root mean square error 8.4.