Evaluating machine learning algorithms for predicting maize yield under conservation agriculture in Eastern and Southern Africa

Abstract

Crop simulation models are widely used as research tools to explore the impact of various technologies and compliment field experimentation. Machine learning (ML) approaches have emerged as promising artificial intelligence alternative and complimentary tools to the commonly used crop production models. The study was designed to answer the following questions: (a) Can machine learning techniques predict maize grain yields under conservation agriculture (CA)? (b) How close can ML algorithms predict maize grain yields under CA-based cropping systems in the highlands and lowlands of Eastern and Southern Africa (ESA)? Machine learning algorithms could predict maize grain yields from conventional and CA-based cropping systems under low and high potential conditions of the ESA region. Linear algorithms (LDA and LR) predicted maize yield more closely to the observed yields compared with nonlinear tools (NB, KNN, CART and SVM) under the conditions of the reported study. However, the KNN algorithm was comparable in its yield prediction to the linear tools tested in this study. Overall, the LDA algorithm was the best tool, and SVM was the worst algorithm in maize yield prediction. Evaluating the performance of different ML algorithms using different criteria is critical in order to get a more robust assessment of the tools before their application in the agriculture sector.