<![CDATA[Global climate change directly affects agricultural productivity and increases uncertainty in crop yield prediction systems. Most machine learning models still rely on historical data that fail to represent extreme climate scenarios. This study proposes a data processing strategy based on generative simulation to evaluate the robustness of crop yield prediction models under synthetic climate perturbations. The Gaussian-based what-if analysis approach was applied to generate synthetic data that preserves the statistical characteristics of the original dataset. The baseline model employed a HistGradientBoostingRegressor, evaluated using R², MAE, RMSE, and Stability Index (SI) metrics. Experimental results achieved an R² of 0.9519, MAE of 1.08 t/ha, and SI values exceeding 0.95 across all simulated rainfall (±15%) and temperature (±2°C) scenarios. The Kolmogorov–Smirnov test confirmed that synthetic data distributions were not significantly different from the original (p > 0.05). These findings demonstrate that Gaussian-based generative simulation effectively enriches agricultural data and enables quantitative sensitivity evaluation of predictive models. The proposed approach aligns with the data-centric AI paradigm and supports the development of resilient, climate-adaptive smart agriculture systems]]>
Copyrights © 2025
