Maize yield predictive models and mobile-based decision support system for smallholder farmers in Africa

Olisah, Chollette; Smith, Lyndon N; Smith, Melvyn L

Abstract

Existing machine learning models for crop yield prediction model environmental data on the assumption that soil variables are unaffected by weather variables and therefore learn their intrinsic features independently. If the focus of crop yield prediction is aimed at supporting smallholder farmers in making farming decisions, then modelling the environmental variables independently might not be informative for the farmer. In this paper, we propose a comprehensive machine learning based crop yield decision support tool for smallholder farmers. It comprises of predictive machine learning models that models the dynamic interactions between environmental variables for predicting crop yield at the level informative to a smallholder farmer. Then, the best model is integrated to a mobile application with farmer education and market access modules to provide the smallholder farmer with a tool that enables him/her to farm smart. From evaluation of our random forest regressor (RFR), extreme gradient boosting regressor (XGBoostR), and multi-layered perceptron regressor (MLPR) using mean squared error (MSE) metric which quantifies the average of the square of the error, the values of 0.0075 t/ha, 0.1416, 0.3031 t/ha were achieved, respectively. This shows that the RFR model best minimizes the error between the predicted and ground truth crop yield value.

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: 17th International Conference on Machine Learning and Data Mining MLDM 2022
• Start Date: Jul 16, 2022
• End Date: Jul 21, 2022
• Deposit Date: Apr 14, 2022
• Keywords: maize; crop yield prediction; machine learning; decision support system; Sub-Saharan Africa
• Public URL: https://uwe-repository.worktribe.com/output/9327477
• Related Public URLs: http://www.mldm.de/