Using Neural Networks to Predict Dye Concentrations in Multiple-Dye Mixtures

Abstract

A neural network is used to predict the concentrations of three dyes from their spectrophotometric absorbances. With this technique, the network achieves a peak of 5% and an average of 2.6% error over a range of dye concentrations of 0 to 0.15 g/l. The neural network is evaluated against two other models based on Beer's law, and performs significantly better over the entire range of concentrations measured.