Using pattern mining to determine fine climatic parameters for maize yield in Benin

Abstract

This study investigates the relationships between Benin's climate and maize production to develop an association rule algorithm for accurate yield prediction. The datasets utilized extend 26 years (1995 to 2020) and include climate and maize yield data from five districts with synoptic weather stations in two agroclimatic zones (Sudanian and Sudano-Guinean). Climate variables were combined with yield using "year" and "districts" to find the association rules. Several techniques were used to determine the correlation between weather parameters and maize yields: support vector machines, K nearest neighbor, artificial neural networks, decision trees, and recurrent neural networks. The most performed method was the decision tree (R2=0.998, mean squared error (MSE)=0.021, and mean absolute error (MAE)=0.0008). This model is difficult to understand, though the frequent pattern growth technique was then applied to the dataset to facilitate the discovery of the rules. The Sudano-Guinean zone exhibits high maize yields for medium minimum and maximum temperature values, rainfall, evapotranspiration, and humidity. In the Sudanian zone, medium minimum and maximum temperatures and maximum humidity levels are associated with high maize yields. The discovered association rules showed that optimizing maize output might be done dependably and effectively.