The Role of Continuous Glucose Monitoring in Automatic Detection of Eating Activities

The automatic detection of eating activities is important for understanding eating patterns and diagnosing eating disorders. In this paper, we investigated the effect of using continuous glucose monitor together with off-the-shelf consumer wearable wristbands in enhancing the classification accuracy of eating and non-eating activities. Our results showed that adding the glucose modality can help reduce false positive rate and achieve a more balanced performance on both eating and non-eating events compared to only using accelerometer and photoplethysmogram (PPG). In the meantime, the effectiveness of the glucose modality was also modulated by the type of machine learning algorithm used.