Abstract—Artificial Intelligence (AI) and its role in precision agriculture are getting away with their greatest potential role in precision agriculture, providing accurate crop yield prediction and proper allocation of resources. In this work, we propose AgroAttenNet, a Convolutional Neural Network (CNN)-Bidirectional Long Short-Term Memory (BiLSTM) based interpretable hybrid deep learning ensemble model with multi-head temporal attention and Random Forest (RF) for the realization and responsible forecasting of groundnut yield. Training of the model was done with 10 years of our most complex multi-modal agro-meteorological and soil data (2013–2023) pertaining to Telangana, India, containing observations like rainfall, temperature, humidity, Normalized Difference Vegetation Index (NDVI), and nutrient profiling across 33 districts. Robust preprocessing and time-aware cross-validation ensured consistency and prevented temporal leakage. AgroAttenNet achieved Root Mean Square Error (RMSE) = 0.426 ± 0.038, Mean Absolute Error (MAE) = 0.331 ± 0.029, and R² = 0.92 ± 0.025 (95% CI: 0.89–0.95). Paired t-tests confirmed that the performance improvements over CNN + BiLSTM and RF baselines were statistically significant (p < 0.001). The attention mechanism identified critical growth phases (weeks 7–10), offering explainable insights for agronomic interventions such as irrigation scheduling and nutrient management. By unifying spatial, temporal, and ensemble learning within a single framework, AgroAttenNet provides a robust and interpretable solution for yield prediction, paving the way for scalable AI-driven decision support systems in semi-arid farming contexts. Future work will extend AgroAttenNet to other crops and regions to evaluate its scalability and generalization capacity.
 
Keywords—deep learning, Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), attention mechanism, ensemble learning, crop yield prediction, precision agriculture, temporal modelling


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).