Answers are grounded in this paper and cite their sources.
Accurate yield prediction under increasing climatic variability is essential for ensuring sustainable vegetable production in Eastern India. Eggplant ( Solanum melongena L.), a major solanaceous crop in West Bengal, is highly sensitive to climatic fluctuations, soil heterogeneity, and moisture stress, particularly during early establishment and reproductive stages. This study developed a robust multi-model machine learning framework to predict district-level brinjal yield across diverse agro-climatic zones using integrated climato-edaphic and hydroclimate-related variables. Multiple base learners were trained and combined through a stacked ensemble approach to capture nonlinear interactions among predictors. Among individual models, Random Forest (rf) performed best, while models like k-Nearest Neighbors (knn) and Elastic Net (glmnet) showed relatively poor performance. Stacking further improved predictive accuracy, with the glmnet-based meta-learner (glmnet_meta) and random forest based meta learner (rf_meta) both achieved high predictive accuracy. Despite comparable performance among top models, glmnet_meta was selected for its structural parsimony and transparency, adhering to the principle of Occam’s razor. Model diagnostics confirmed reliability, with near-normal residuals and no heteroscedasticity. SHAP-based analysis indicated that soil attributes were the primary predictive contributors of yield variability (44.9%), followed by weather variables (29.2%), hydroclimatic indices (22.1%) and engineered biophysical interactions (3.7%). Soil water holding capacity (WHC) and nitrogen (N) emerged as the most critical determinants for yield prediction. Temperature and rainfall extremes also showed significant effects, with distinct spatial variability across agro-climatic zones. Furthermore, spatial validation approaches confirmed that while the model is highly stable for temporal forecasting, it is subject to spatial boundary constraints. Overall, the proposed framework integrates predictive accuracy with interpretability, offering a robust understanding about yield prediction, predictive contributors and methodological proof-of-concept validated under West Bengal conditions.
No takes yet. Share an insight, caveat, or question.
Sengupta et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: