Evaluating the efficiency of NDVI and climatic data in maize harvest prediction using machine learning

Abstract

ABSTRACT Accurate anticipation of the maize harvest date is important in the agricultural market, as it ensures the sustainability of food production in response to the increasing global demand for food. This paper proposes a predictive model to determine the optimal harvest time in maize plots using the Normalised Difference Vegetation Index (NDVI) and climatological data. These variables were oversampled and used to train various models, including Random Forest (RF), Gradient Boosting Machine (GBM), Light Gradient Boosting Machine (LGBM), Extreme Gradient Boosting Machine (XGBoost), CatBoost and Support Vector Machine (SVM). Bayesian optimisation has been used to find the best hyperparameters and Shapley values to identify the variables that exert the most significant influence on the prediction in each model instance. As a result of this approach, a model with an accuracy of 92.1% and an Area Under the Curve (AUC) of 0.935 was developed. The variables that determined these results were atmospheric pressure, mean temperature, precipitation, NDVI, and precipitation.