A Machine Learning Approach to Crop Yield Forecasting Using Meteorological and Pesticide Data

Crop yield forecasting is an essential parameter in nutrition security and in planning the utilization and productivity of agricultural resources. Using traditional models, some problems have emerged with complexity and interdependency of meteorological parameters, pesticides and yields. The new machine learning model is the accumulation of meteorological data and pesticide data for predicting the crop yield. Random Forest and Gradient Boosting are two complicated ensemble learning algorithms of the framework, as well as the feature engineering process that reflects input-output interaction. The results indicated an accuracy of 95.1 % and a value of R-square was 0.93 and the error rate was 1.20% lower of the other models. Moreover, the model provides useful advice, the guideline for efficient pesticide use and risk assessment in climate change. These result indicate the applicability of the proposed framework in enhancing the yield prediction and thereby the decision making process in agriculture.