Harendra Singh Negi
Department of Computer Science and Engineering, Graphic Era (Deemed to be University), Dehradun, Uttarakhand, India.

Sushil Chandra Dimri
Department of Computer Science and Engineering, Graphic Era (Deemed to be University), Dehradun, Uttarakhand, India.

DOI https://doi.org/10.33889/IJMEMS.2026.11.3.057

  ;

Abstract

As the issue of agricultural sustainability has continued to increase, there has been a need to use data based solutions to improve agricultural productivity. This paper proposes a machine learning system combining Random Forest and XGBoost to combine prediction-forecasting crop yield and classification of crop type rice and wheat in Indian state of Uttarakhand. The model is tested using a library of 6, 000 samples containing 12 soil and climatic characteristics and measured on regression and classification quality. The hybrid ensemble with hyperparameter optimization and verified on the basis of 10-fold cross-validation performed better than single base learners in all measures. It achieved a classification accuracy of 96.3 and R2 = 0.927. Statistically significant developments that were formed using paired t-tests were set at p = 0.05. The SHAP and ablation analysis found out nitrogen, rainfall, and pH as the most influential features. The forecasted framework provides a better generalizability, interpretability, and computational effectiveness, which is appropriate to be applied in the designs of real life in precision agronomy. The new result is novel, interpretable, and high-performative to crop yield intelligence in data-scarce areas and provides a contribution to this study.

Keywords- Ablation, PCA, Hyperparameter, SHAP, Statistical validation.

Citation

Negi, H. S. & Dimri, S. C (2026). SHAP-Backed Hybrid Ensemble Model for Rice and Wheat Forecasting in Data-Scarce Environments. International Journal of Mathematical, Engineering and Management Sciences, 11(3), 1395-1423. https://doi.org/10.33889/IJMEMS.2026.11.3.057.