Agricultural Crop Yield Prediction Using Machine Learning

Abstract

Crop yield prediction is addressed through machine learning. Two predictor variables were used: hectares harvested, and production in tons. For the first case, the best model was a dense neural network (DNN) architecture, with a MSE of 0.0081, followed by Random Forest (RF) with an MSE of 0.0104, decision trees (AD) with 0.0168, and finally vector support machines (SVM) with 0.0328. When production in tons was predicted, the best model was RF with a MSE of 0.0550, followed by AD with 0.1418, DNN with 0.1489, and finally SVM with 0.3420. The statistical test of significant difference showed that there is no such difference between the performance of the models when the variable hectares harvested is predicted, but in the case of production in tons, where the predictive capacity of RF was approximately 95%.

Keywords

• agricultura
• investigación agrícola
• inteligencia artificial.