A MACHINE LEARNING REGRESSION MODEL FOR PREDICTING THE OCCURRENCE OF AFLATOXIN IN PEANUT FIELDS USING ENVIRONMENTAL AND FIELD-DEPENDENT VARIABLES

This study was focused on studying and developing machine learning techniques to quantify the relationship between aflatoxin production in peanut plants and field parameters, and delineate a relationship between the same. The study was conducted on rainfed farmers’ fields in South-Georgia to replicate the conditions of a dry cropland. The study found that soil temperature, soil texture and weather parameters like solar radiation, air temperature, and VPD can be used to explain the variation in aflatoxin production using a random forest regression model, paired with feature engineering methods (Recursive Feature Elimination) and cross-validation techniques, with an accuracy of 27% and an RMSE of 0.65.