ARCHIVES

Accurate prediction of crop yields is critical for planning in the agricultural sector. Although many prediction models have been developed in recent years with improved accuracy, there is a lack of investigation into the contribution of each group of agricultural features to the predictions of crop yield. Additionally, many of the best-performing models rely on data from remote sensors in agricultural environments, limiting their applicability to those environments. To investigate the contribution of agricultural variables to the prediction of crop yield, methods including data ablation and Explainable Artificial Intelligence (XAI) were used. Datasets comprising agricultural variables such as climate, soil, and management variables were collected from publicly available databases. Machine learning models were developed using these datasets. The influence of each group of agricultural variables was determined using data ablation. Additionally, XAI methods using SHapley Additive exPlanations (SHAP) were used to further determine the importance of each group of agricultural variables. The results indicate that the variables related to climate contribute the most to the performance of the prediction models. The variables related to soil and management contribute to the accuracy of the models when incorporated with the climate-related variables. The SHAP results confirm the importance of the climatic variables and demonstrate interactions among the different agricultural variables. The results of this study show that using data ablation and XAI methods provides an understanding of the importance of the features in prediction models, beyond that provided by model accuracy rates. The insights from this study can contribute to the development of machine learning models for the prediction of crop yields that are interpretable, accurate, and suited for use in agricultural settings.