Performance Analysis of Eating Spotting using EMG Eyeglasses

Abstract

We spot eating using data recorded with Electromyogram-monitoring eyeglasses in free-living. We apply overlapping and non-overlapping sliding windows to the data and use one-class classification for spotting. We evaluate the spotting timing errors at eating start and eating end. Results show that best F1-score may not provide best timing performance. Timing errors for small windows, e.g. 5 s, were larger than 100 s.