Application of XGBoost Regression in Maize Yield Prediction

Artificial Intelligence (AI) is the human-like intelligence imbued in machines so that they can perform tasks that normally require human intelligence. Machine learning is an AI technique which carries on the concepts of predictive analytics with one important distinction: the AI system can make assumptions, test hypotheses, and learn independently. XGBoost, Extreme gradient boosting, is a popular machine-learning library for regression tasks. It implements the gradient-boosting decision tree algorithm, which combines several feeble decision trees to produce a robust predictive model. In Boosted Trees, boosting is the process of transforming poor learners into strong learners. It is an ensemble method; a weak learner is a classifier with a low correlation with classification, whereas a strong learner has a high correlation. Maize is a staple food in Kenya and having it in sufficient amounts in the country assures the farmers' food security and economic stability. Crop yield measures the seeds or grains produced by a particular plot of land. Typically, it is expressed in kilograms per hectare, bushels per acre, or sacks per acre. This study predicted maize yield in Uasin Gishu, a county in Kenya, using XGBOOST regression algorithm of machine learning. The regression model used the mixed-methods research design, the survey employed well-structured questionnaires comprising of quantitative and qualitative variables, directly administered to selected representative farmers from 30 clustered wards. The questionnaire comprised 30 variables related to maize production from 900 randomly selected maize farmers distributed across 30 wards. XGBOOST machine learning regression model was fitted, and it could predict maize yield and identify the top features or variables that affect maize yield. The model was evaluated using regression metrics Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE), which values were 0.4563, 0.2082, 25.2700 and 0.3532, respectively. This algorithm was recommended for maize yield prediction.