A novel method for predicting cadmium concentration in rice grain using genetic algorithm and back-propagation neural network based on soil properties.

Heavy metal pollution is a global ecological safety issue, especially in crops, where it directly threatens regional ecological security and human health. In this study, the back-propagation (BP) neural network optimized by the genetic algorithm (GA) was used to predict the concentration of cadmium (Cd) in rice grain based on influencing factors. As an intelligent information processing system, the GA-BP neural network could learn the laws of Cd movement in the soil-crop system through its own training and use the soil properties to predict the concentration of Cd in grain with high accuracy. The total soil Cd concentration, clay content, Ni concentration, cation exchange capacity (CEC), organic matter (OM), and pH have important impacts and interactions on Cd concentration in rice grain were selected as input factors of the prediction model based on Pearson's correlation analysis and GeoDetector. By using GA to optimize the initial weight, the prediction accuracy of the GA-BP neural network model was optimal compared with the BP neural network model and multiple regression analysis. Based on the Cd concentration predicted in grain by the model, human exposure and health risk can be assessed quickly, enabling measures to be taken in time to reduce the transfer of Cd from soil to the food chain.