Drying Kinetics of Green Tomatoes (Physalis ixocarpa) Integrating Mathematical Models

This study explores the application of mathematical models and machine learning techniques to describe and predict food drying kinetics during dehydration. A total of 14 mathematical models were evaluated, including Newton, Page, Modified Page, Henderson and Pabis, Logarithmic, Two-Terms, Wang and Singh, Thomson, Approximate Diffusion, Werma, Modified Henderson and Pabis, Two-Term Exponential, Simplified Fick’s Diffusion, and Modified Page, to determine their fit to the drying kinetics of green tomatoes under controlled temperature and airflow conditions. The results revealed that the Werma, Two-Term Exponential, and Modified Page models provided the best fits, with R² values exceeding 0.998 and RMSE values ranging from 0.01268 to 0.01345, demonstrating their high accuracy. Other models, such as Approximate Diffusion and standard Page, also showed good performance, with R² values close to 0.998 and RMSE values below 0.015. In contrast, the Thomson and Wang and Singh models displayed notable discrepancies, with lower R² values of 0.9647 and 0.9532, respectively, and higher RMSE values reaching up to 0.08284 for Wang and Singh. Following the identification of the most accurate models, a Convolutional Neural Network (CNN) combined with Long Short-Term Memory (LSTM) was implemented to optimize the prediction of the drying process. The integration of the CNN-LSTM model significantly improved the prediction of moisture content over time. The Two-Term Exponential model was found to provide the best fit, while the CNN-LSTM model enhanced the predictive capabilities of drying kinetics. These findings highlight the potential of combining traditional mathematical models with advanced machine learning techniques to optimize the food dehydration process.