Predicting dielectric properties of fruit juices at 915 and 2450 MHz using machine learning and physicochemical measurements

Microwave-assisted thermal processing can provide superior quality for fruit-based products when compared to conventional thermal processing. Understanding the temperature-dependent dielectric properties of liquid foods is needed for the analysis and optimization of the microwave applicator chamber since they govern the heating rate and temperature distribution. While literature offers correlations for specific products, there is a scarcity of methods capable of accommodating variability in composition or predicting behavior for broader product groups. In this study, we measured the dielectric properties (dielectric constant and loss factor) of eight fruit juices (passion fruit, melon, pineapple, cashew, orange, lemon, acerola, and guava) using an open-ended coaxial-line technique for temperatures ranging from 5 to 90 °C at commercial frequencies of 915 and 2450 MHz, alongside electrical conductivity. These properties were successfully correlated with the temperature for each individual juice; then, machine learning techniques (random forest, gradient boosting machine, and multilayer perceptron) were used to predict the properties of this diverse group of eight juices based on various physicochemical measurements. These techniques revealed temperature and electrical conductivity as the most critical predictors, while total solids, pH, acidity, ashes, and select color parameters also emerged as significant variables. These findings demonstrate that the integration of physicochemical analyses with machine learning tools offers an objective approach to correlate and predict dielectric properties for a group of food products, facilitating adjustments in product composition without additional measurements, thus enhancing the efficiency and accuracy of microwave-assisted thermal processing simulations and optimizations.