Crop Yield Prediction using Gradient Boosting Regression

Abstract

Achieving greater crop yields remains a pressing challenge for both farmers and governments. This research examines the use and implementation of Gradient Boosting Regression in predicting crop yields for numerous districts in France. XGBoost, an efficient, optimized and flexible distributed gradient boosting library was used. Agricultural data was sourced from the CLAND Institute’s ‘Crop Data Challenge 2018’, which contains approximately 38 years of maize data compiled by various departments from the months of January to September from regions in France. Average monthly climatic parameters such as evapotranspiration, maximum and minimum temperature, cumulative precipitation, yield anomaly, solar radiation levels and an irrigation coefficient were used as input variables. The best result obtained was an RMSE value of 0.755 and MAE error of 0.54 was obtained using a tuned XGBoost Regressor trained on original variables. This paper aims to compare various regression techniques in order to improve yield prediction thus giving farmers a chance to improve their cultivation with better insights as well as enabling them to harness the power of predictive analytics.