Classification of Eating Behaviors in Unconstrained Environments

Obesity and its numerous devastating consequences are on the rise globally. While widespread tactics to fight against obesity often focus on healthy eating, how the food is consumed is oftentimes overlooked even though convincing evidence attests that merely eating slowly and properly chewing one’s meal significantly reduces obesity. This research introduces a method that recognizes common human actions during mealtime—namely, food chewing, food swallowing, drink swallowing, and talking. The proposed system is unobtrusive. It uses a cheap and small bone conduction microphone to collect intra-body sound and a smartphone that provides feedback in real-time. Our proposed approach achieves similar performances (Accuracy = 97.5%, Specificity = 98.0%, Precision = 83.8%, Recall = 91.7%, \(F_1\) score = 87.2%, and MCC = 0.85) as those achieved by the most recent state of the art models even though our system uses modest machine learning models.