Application of Ensemble Machine Learning Techniques in Yield Predictions of Major and Commercial Crops

Farming activities and crop productivity has been greatly affected by climate change, soil fertility, and the availability of planting areas in recent decades. Early estimation of yields and their quality are the prime requirements in the global food market which depend on several input parameters. Predicting crop yields gives the farmers an insight to decide on the crop production rate and cultivate suitable crops for the given climatic conditions. The present work aims at estimating yields of major and cash crops of Karnataka using climate and crop-related data through four ensemble regression approaches. The feature importance concept is incorporated that describes which features contribute most to the prediction results in all the models that help in better data interpretation. The performance metrics such as MSE, MAE, and R² were adapted to measure the accuracy. The extreme gradient boosting regressor was found to deliver the best performance (R² of 0.999) while incurring the lowest computational cost among the four ensemble regressor models. The results showed that ensemble regression methods can be effectively used for yield predictions when there is a medium-sized dataset.