Optimized Neural Network for Instant Coffee Classification through an Electronic Nose

Abstract

Flavor is one of the most important features of food, especially of coffee. The evaluation of this sensory feature is complex yet indispensable in quality control of instant coffees. In this work, an artificial neural network (ANN) was developed for instant coffee classification based on an electronic nose (EN) aroma profile. To this purpose, a hybrid algorithm was developed, containing: bootstrap resample methodology; factorial design and sequential simplex optimization to tune network parameters; an ensemble multilayer perceptron (MLP) trained with backpropagation for coffee classification; and causal index procedure for knowledge extraction from the trained ANN. The produced neural network classifier correctly recognizes 100% of coffees studied. Furthermore, the causal index employment allowed inference of some rules on how the coffees were separated according to the sensors available in EN. The results indicate that the applied methodology is a promising tool for instant coffee quality control.