Machine Learning powered E-Nose for Liquid Food Classification

Monitoring food intake provides valuable insight to the individual dietary routine. Self-reporting of the food intake on daily basis is viewed as a tedious process, leading to under-reporting, which affects the overall evaluation on the food intake. Hence a technologically aided autonomous food monitoring and reporting solution is much needed for effective management and characterizing individual well-being. Wearable sensors, and camera based solutions have shown to effectively track food of solid form. However, tracking liquid food item through these set of sensors is likely to fail due to minimal distinguishing features among the consumable liquid food items. Hence a novel electronic-nose (e-nose) system encasing four gas sensor channels is designed to detect the liquid food intake precisely. This solution along with the existing wearable sensors is expected to categorize various food items either in solid and liquid form and aids in autonomously managing the eating habits of the individual. This work presents a novel E-Nose portable platform comprising of a glass container with a gas sensor interfaced to the micro-controller for signal acquisition. The machine-learning (ML) model is trained for large dataset consisting of temporal responses from the four sensory channels. The acquired sensory response for all the five liquids were studied in two-dimensional (2D) feature space to showcase the classification capability. All the three ML models including Artificial Neural Network, Random Forest, and K-Nearest Neighbours achieved high accuracy which is attributed to the distinguishing ability of the sensors response for the five liquid food items under investigation. This work forms a fundamental basis to further characterize different concentration of ingredients within a single and multiple such liquids when served on the table.