Mathematical and Deep Learning Modelling of the Raspberries Drying Kinetics

This study investigates the modeling of raspberry drying kinetics using both mathematical and deep learning approaches. Red raspberry (Rubus idaeus L.) is a highly cultivated and exported fruit in Serbia, with an annual production exceeding 60,000 tons. The study aims to enhance the understanding and prediction of drying kinetics, which is crucial for optimizing drying processes and ensuring product quality. The experimental setup involved the convective drying of raspberries at different temperatures and air flow speeds, with the data collected and processed automatically. Twelve mathematical models were analyzed, with the Newton and Aghbashlo models showing the best performance based on R2 and χ2 values. Additionally, two machine learning models, Artificial Neural Networks (ANN) and Recurrent Neural Networks (RNN), were developed and optimized by varying hyperparameters such as the number of hidden layers, neurons, learning rate, and epochs. The ANN model achieved an R2 value of 0.99923 and an RMSE of 0.00864, while the RNN model achieved an R2 value of 0.9966 and an RMSE of 0.00186. Both models demonstrated excellent predictive capabilities, with minimal errors compared to the experimental data. The study concludes that the selected ANN and RNN models are highly effective for predicting raspberry drying kinetics and can replace traditional semi-empirical models. Future research will focus on further refinement of these models and application to other types of biological materials to achieve even higher accuracy and broader applicability.