Prediction of the Initial Freezing Temperature of Exotic Fruits Grown in Peru Using Artificial Neural Networks

This study aimed to develop and evaluate different artificial neural network (ANN) architectures to predict the initial freezing temperature (Tf) of exotic Peruvian fruits. The physicochemical properties of 17 commercial fruits: moisture content (%H), soluble solids (°Brix), pH, titratable acidity (%), and Tf were experimentally determined and used to train various ANN models and construct a nonlinear empirical model. The most accurate ANN consisted of three hidden layers with 20 neurons each and sigmoid activation functions, achieving a maximum absolute deviation (AD) of ±0.18°C. The proposed empirical model (Tf = −0.21°Brix^0.98 %H^1.57) also showed good predictive performance (AD = ±0.37°C). Both models were validated using exotic fruits cultivated in Peru (pitahaya, aguaje, cherimoya, camu camu, and sanky), with the ANN achieving a maximum AD of ±0.24°C and a nonlinear model with a maximum AD of ±0.59°C. In conclusion, the developed ANN demonstrated high accuracy in predicting Tf, outperformed previously reported models in the literature, and represents a promising tool for applications in tropical fruit processing and preservation.