Combining miRNA concentrations and optimized machine-learning techniques: An effort for the tomato storage quality assessment in the agriculture 4.0 framework

Abstract

Fruits’ cold storage lead to an increase or decrease in the concentration (expression) of several miRNAs in their intracellular structure. Moreover, research has shown that conventional machine-learning methods do not exert enough performance in predicting treatments applied to plants by having miRNA concentrations. In this work, using basic machine-learning methods and their optimization via meta-heuristic algorithms, the storage period, storage temperature, and mechanical loading during storage in tomatoes have been predicted by having miRNA concentrations as model inputs. As expected, the results showed rather poor values of the coefficient of determination (R²) of the artificial neural network (ANN), support vector machine (SVM), and random forest (RF) with pre-adjusted values for their hyperparameters. However, the RF, with hyperparameters optimized by the genetic algorithm, was able to improve the R² values of the prediction of storage temperature and period to 0.96 and 0.89. The maximum performance of predicting the mechanical loading on the fruits (R² = 0.91) was obtained by combining the RF with the particle swarm optimization. Also, feature selection results showed that miRNA1917, miRNA172, and miRNA156, as inputs to the optimized RF model could predict the storage temperature, storage period, and mechanical loading on the fruits with R² values of 0.94, 0.93, and 0.93, respectively. As a result, to use smart sensing platforms to detect the storage quality of agricultural products, only a limited number of miRNAs is required to be measured, which reduces the redundancy of the database and also reduces the costs of experiments. In addition, this feature selection scheme reveals the role of some miRNA compounds in the process of fruit response to stress during storage. This study is an effort to move along the Sustainable Agriculture 4.0 b y introducing a reliable method to predict fruit storage conditions for applying possible treatments to reduce post-harvest loss.