EUVIMEDEUVIMED entdecken
Crossref

Hybrid Improved Stacking over Tabular Temporal Features with Blockchain-Certified Data: Millet Yield Prediction and Explainability

Pape ElHadji Abdoulaye Gueye, Cherif Bachir Deme, Diery Ngom, Adrien Basse

Information Dynamics and Applications · 30.12.2025 · 4 (4) · 238-256

DOI: 10.56578/ida040405 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Information Dynamics and Applications, online erschienen am 30.12.2025. Verfasst von Pape ElHadji Abdoulaye Gueye und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate crop yield prediction is essential for food security planning in developing countries.However, real-world deployments remain challenging due to limited imagery availability, heterogeneous tabular data, and concerns regarding data reliability.This paper proposes a tabular-only temporal deep learning framework enhanced with a blockchain-based data provenance layer for millet yield prediction in Senegal.The proposed model embeds per-timestep agroecological features using a multilayer perceptron (MLP), captures temporal dependencies through a bidirectional Long Short-Term Memory (BiLSTM) network, and integrates a hybrid improved stacking strategy by incorporating predictions from classical machine learning models, including Random Forest, XGBoost, LightGBM, and CatBoost.Unlike conventional stacking approaches, these predictions are injected directly into the temporal representation at the final timestep, thereby improving generalization and calibration performance.To ensure data integrity and traceability, a blockchain-inspired certification mechanism is introduced.This mechanism relies on canonicalization, SHA-256 hashing, and HMAC-based signatures of zone-year records.Experimental results demonstrate that the proposed approach achieves strong predictive performance (MAE ≈ 0.074, RMSE ≈ 0.101, R 2 ≈ 0.946), outperforming baseline models.A comprehensive evaluation framework is employed, including cross-validation, statistical significance testing, and explainability analysis using SHAP, LIME, and gradient-based saliency methods.Results indicate that while performance improvements are significant under static evaluation settings, they are less consistent under temporal cross-validation, highlighting the importance of robust evaluation protocols.Overall, the proposed framework provides a practical, auditable, and high-performing solution for yield prediction in data-scarce environments, combining predictive accuracy with data trustworthiness.

Zitieren

APA

Gueye, P. E. A., Deme, C. B., Ngom, D., & Basse, A. (2025). Hybrid Improved Stacking over Tabular Temporal Features with Blockchain-Certified Data: Millet Yield Prediction and Explainability. Information Dynamics and Applications, 4(4), 238-256. https://doi.org/10.56578/ida040405

Vancouver

Gueye PEA, Deme CB, Ngom D, Basse A. Hybrid Improved Stacking over Tabular Temporal Features with Blockchain-Certified Data: Millet Yield Prediction and Explainability. Information Dynamics and Applications. 2025;4(4):238-256. doi:10.56578/ida040405

BibTeX
@article{Gueye2025,
  author = {Gueye, Pape ElHadji Abdoulaye and Deme, Cherif Bachir and Ngom, Diery and Basse, Adrien},
  title = {Hybrid Improved Stacking over Tabular Temporal Features with Blockchain-Certified Data: Millet Yield Prediction and Explainability},
  journal = {Information Dynamics and Applications},
  year = {2025},
  volume = {4},
  number = {4},
  pages = {238--256},
  doi = {10.56578/ida040405}
}
Kostenlos starten →