Assessing a Machine Learning Model for Predicting Maize Grain Yield Based on Chlorophyll Content and Vegetation Indices

This study investigates maize productivity using remote sensing data, vegetation indices, and machine learning techniques. Utilizing a multi-temporal dataset of 17,100 images captured by unmanned aerial vehicles (UAVs) across three crop cycles, we extract essential vegetation indices. These indices, along with ground-truth data on maize yields and chlorophyll content, serve as inputs for twelve machine learning models, including CatBoost, random forest, and XGBoost regressions. Rigorous evaluation, including RMSE, MAE, and R
²
, assesses each model’s capacity to predict maize yields.Key vegetation indices with strong positive correlations to maize yields emerge, offering predictive markers for precision agriculture. The best-performing model, linear regression, achieves an impressive R
²
of 0.77, while the worst-performing model, support vector regression, exhibits a negative R
²
of −0.74.These results emphasize the significance of model selection in maize yield prediction, guiding agricultural decision-making. Future research can delve deeper into enhancing agricultural practices and food security through refined yield prediction dynamics.