Utilizing Artificial Neural Networks for Predictive Modeling Physicochemical Attributes in Maltodextrin-Coated Grapes with Potassium Carbonate and Pyracantha Extract in Storage

Artificial neural networks (ANN) are a nondestructive method for estimating fruit and vegetable shelf life and quality attributes. This research used artificial neural networks to model a storage process for fruit grapes (Vitis vinifera cv. Rishbaba) coated with maltodextrin, including different levels of potassium nanocarbonate (0-2%) and pyracantha extract (0-1.5%). After applying these coatings, the fruits were stored for 60 days in cold storage (-1 °C), with a relative humidity of 90%. Measurements considered weight loss percentage, titrable acidity (TA), pH, texture firmness, color index (a), and general fruit acceptance. Artificial neural networks predicted changes in fruits during the storage process. By examining different networks, the feedforward backpropagation network had 3-10-6 topologies with a coefficient of determination (R2) greater than 0.988 and a mean square error (MSE) less than 0.005. With a hyperbolic sigmoid tangent activation function, a resilient learning pattern and 1000 learning process were determined as the best neural method. On the other hand, the results of the optimized models showed that this model had the highest and lowest accuracy for predicting the weight loss percentage (R2 = 0.9975) and a (R2 = 0.5671) of the samples, respectively.