Predicting the Rate of Aflatoxin Contamination in the White Corn Value Chain

With the digital revolution, computer data is a resource of inestimable value. Businesses use them to understand consumer behavior, make informed decisions, and anticipate market trends. Governments rely on data to develop effective policies, monitor public health and manage resources. In this work, digital data collected in the agriculture sector allowed us to compare different machine learning models to predict aflatoxin infection levels in white corn crops. To do so, we use some qualitative and quantitative variables collected on the field, during a previous work. The compared methods are linear regression, random forests, artificial neural networks and support vector regression. The results of the analysis indicate that the random forest regression model stood out as the most effective in predicting aflatoxin infection levels. It posted an RMSE of 0.14 on the training set and 0.29 on the test set, accompanied by a coefficient of determination of 0.81, demonstrating its robustness on both data sets. This performance can be attributed to the ability of random forests to capture the complex and non-linear relationships between maize traits and aflatoxin levels. Evaluating the models on a separate test set confirmed their generalizability, that is, their ability to maintain accuracy with new data. This result constitutes a promising tool for actors in the agricultural sector, providing valuable information for risk management and strategic decision-making aimed at reducing consumer exposure to aflatoxin, thus contributing to the improvement of food safety and public health.