Comparison of Gaussian process regression, artificial neural network, and response surface methodology modeling approaches for predicting drying time of mosambi (Citrus limetta) peel

Abstract

In this study, drying kinetics of mosambi peel was studied. The effect of three variables viz., temperature (50, 58, 70, 82, and 90\u2009°C), salt concentration (2, 5, 8, and 10%), and thickness of drying bed (1, 2, 3, and 4 mm) on drying time was determined by the central composite design. Applicability of Gaussian process regression (GPR)‐based approach for modeling drying kinetics was analyzed. GPR‐based model was compared with the commonly used approaches like artificial neural network (ANN) and response surface methodology (RSM). The models were validated by comparing model simulations with observed values for unseen data. The models were compared based on performance indices like coefficient of determination, mean square error, root mean square error (RMSE), model predictive error, mean average deviation, goodness of fit, and chi‐square analysis. All the three models fit both seen and unseen data excellently. RMSE, mean average deviation, and model predictive error for the unseen data of the GPR‐based model were minimum (0.191, 0.285, 6.8%, respectively) followed by ANN (0.35, 0.298, 7.2%, respectively) and RSM (1.162, 0.905, 32.0%, respectively). PRACTICAL APPLICATIONS: Modeling of the drying process is very important for control of industrial dryers. Several attempts have been made to model drying kinetics using RSM and ANN. This study shows the efficiency of GPR in modeling the drying kinetics of mosambi peel for the first time. GPR‐based models were found to be a better alternative to RSM and ANN. This will help in developing more accurate models and increase the efficiency of drying.