Predictive modeling of mass transfer kinetics in convective drying of fruit slices using dragonfly-optimized SVM and ANN

Abstract

Determining optimal drying parameters is critical in the chemical and food industries to mitigate energy waste and ensure product quality. This study focuses on modeling moisture content and mass transfer kinetics during the convective drying of fruit slices using artificial intelligence techniques and comparing their performance with that of conventional models. We evaluated the performance of four distinct machine learning models: two artificial neural networks (ANN1 and ANN2), a standard support vector machine (SVM), and a hybrid SVM optimized by the Dragonfly Algorithm (DA-SVM). The models were trained and validated using a comprehensive dataset of 7,760 experimental data points collected from the scientific literature and representing moisture-content drying curves over time under various operating conditions. The optimized two-hidden-layer ANN2 model demonstrated superior performance compared with the other evaluated models, showing excellent agreement between the predicted and experimental values with R ²  = 0.9987, RMSE = 0.0787, and AARD = 2.8261 %. The predictive performance of ANN2 was also compared with that of classical empirical formulations, including the Page, Henderson–Pabis, Lewis, Logarithmic, and Wang–Singh models. To quantify the influence of each input variable on the model predictions, a sensitivity analysis based on the generalized Garson algorithm was performed. To enhance practical applicability, a user-friendly graphical user interface (GUI) was developed to facilitate rapid simulation of moisture-content evolution and drying kinetics. This approach provides a robust and reliable tool for improving the prediction, control, and understanding of the complex physicochemical phenomena in fruit dehydration processes.