Quantification of Spice Mixture Compositions by Electronic Nose: Part I. Experimental Design and Data Analysis Using Neural Networks

A quantitative procedure was developed to predict the composition of ternary ground spice mixtures using an electronic nose. Basil, cinnamon, and garlic were mixed in different compositions and presented to an enose. Nineteen training mixtures were used to build predictive models. Model performance was tested using 5 other mixtures. Three neural network structures—multilayer perceptron (MLP), MLP using principal component analysis as a preprocessor (PCA-MLP), and the time-delay neural network (TDNN)—were used for predictive model building. All 3 neural network models predicted the testing mixtures’ compositions with a mean square error (MSE) equal or less than 0.0051 (in a fraction domain where sum of fractions = 1). The TDNN provided the smallest MSE.