Prediction of Tomato Postharvest Quality Using Artificial Neural Networks (ANN): A Case Study on Edible Coating Treatments

This study examined the physicochemical, antioxidant, and microbiological properties of tomatoes coated with gum arabic (G), chitosan (CH), a gum arabic–chitosan blend (G+CH), and a composite enriched with Saussurea costus extract (X). Coating pH ranged from 4.68 (CH) to 5.80 (G), while viscosity increased from 0.019 Pa·s (G) to 0.044 Pa·s (X). Transparency was lowest in X (68.41%), and water vapor permeability (WVP) was lowest in X (3.41 g·mm/m2·day·kPa), indicating superior barrier performance. After 15 days at 4 °C, weight loss was significantly reduced in G+CH (4.66%) and X (4.83%) compared to control (7.64%). Firmness was best preserved in G+CH (80.00 N), and X retained the highest titratable acidity (0.46%), ascorbic acid (18.21 mg/100 g), and total phenolic content (24.29 mg GAE/100 g). Microbial growth in X remained below 7 log CFU/g by day 30. A multilayer perceptron (MLP)-based Artificial Neural Network (ANN) model was developed using five input neurons (coating type and storage time), one hidden layer of 10 neurons, and six output neurons (WL, FM, TA, TSS, AAC, TPC). The model achieved excellent performance with R2 = 0.961–0.991 and RMSE = 0.019–0.811. The ANN model (R2 > 0.95) provides a robust predictive tool for optimizing edible coating formulations.