Prediction of process cheese instrumental texture and melting characteristics using dielectric spectroscopy and chemometrics.

This study evaluated the potentiality of dielectric spectroscopy as a tool to predict the functional properties of process cheese. Dielectric properties of process cheese were collected over the frequency range 0.2 to 3.2GHz at 25°C. Dielectric spectra of process cheese were collected using a high-temperature, open-ended dielectric probe connected to a vector network analyzer. The present study was conducted using 2 sets of commercial process cheese formulations and a set of specially formulated process cheese. For the all the process cheese samples analyzed, a decrease in dielectric constant and dielectric loss factor was observed as the incident frequency increased. Partial least square regression (PLSR) and multilayer perceptron neural network models were developed using the dielectric spectra of process cheese to predict the hardness (gf), melting point (°C), and modified Schreiber melt diameter (mm) of process cheese. The prediction models were validated using the full cross-validation method. The ratio of prediction error to deviation was greater than 2 for melt diameter and hardness, indicating a good practical utility of the PLSR prediction models. The predictability of multilayer perceptron neural network was less than the PLSR models and could be due to the small number of training samples in the data sets. Dielectric spectroscopy coupled with PLSR could be a useful tool for the nondestructive measurement of functional properties of process cheese.