Robust Meta-Knowledge-Informed Model for Precise Crop Yield Estimation Using Transferable Adaptive Method

Precise crop yield prediction is of utmost importance for ensuring the sustainability of agriculture. The current yield prediction models suffer from several limitations, including inadequate hyperparameter tuning, robustness, and limited transferability of meta-knowledge. To address these challenges, this study introduces a framework that utilizes the private dataset of Sindh Crops and incorporates meta-knowledge by integrating Bayesian Optimization and XGBoost Regressor, which effectively eliminates the requirement for manual tuning of hyperparameters. The framework achieves remarkable performance through rigorous experimentation with two surrogate types: (1) Gaussian Process (gp) and (2) Random Forest (referred to as prf) of the transferable topov3 model., as evidenced by high R2(0.8191 for gp and 0.7812 for prf) values. The results., encompassing metrics like MSE., MAE., scaled MSE., and scaled MAE., underscore the framework's significance in advancing yield estimation and benefiting the agricultural stakeholders.