MACHINE LEARNING-DRIVEN PADDY YIELD PREDICTION: COMPARATIVE EVALUATION OF BASELINE VS. ENSEMBLE MODELS IN UDHAM SINGH NAGAR, UTTARAKHAND
Rice is a cornerstone of food security in India, supporting millions of livelihoods and the national economy. However, erratic climate patterns are making paddy yields increasingly unpredictable. This study develops a machine learning framework for rice yield prediction in Udham Singh Nagar district, Uttarakhand, by integrating weather, soil, and crop data. Among baseline classifiers, CatBoost performed best with 80.85% accuracy and a ROC-AUC of 0.90. To further enhance performance, Optuna-tuned CatBoost, XGBoost, and LightGBM models were combined into hybrid ensembles. The Weighted Hard Voting classifier, giving higher weight to CatBoost ([3,1,1]), achieved the highest accuracy of 97.37%, followed by Stacking (95.6%) and Soft Voting ensembles (up to 96%). These results were supported by strong ROC-AUC scores. Overall, the study shows that carefully optimized ensemble models can significantly improve yield prediction accuracy, offering a practical tool for more precise and sustainable rice farming in climate-sensitive regions of India.
Сведения
- Опубликовано
- 2026 · Т. 23, № 2, С. 290-311
- Язык
- Английский
- ISSN
- 1998-6688 · 2959-8109
- Издатель
- Kazakh-British Technical University
- Аффилиации
- Soban Singh Jeena University
Kumaun University
Graphic Era Hill University