DOI: 10.11118/ecos2026_010

PREDICTIVE ANALYTICS IN NEPALESE AGRI-CULTURE

Sandesh Kumar Kafle¹, František Dařena¹

¹ Department of Informatics, Faculty of Business and Economics, Mendel University in Brno, Zemědělská 1, 613 00 Brno, Czech Republic

This study addresses climate vulnerability and the lack of data-driven tools in Nepalese agriculture by building predictive crop-yield models for 14 districts in the Koshi Province. Applying a data-science pipeline to historical agricultural records and NASA POWER agroclimatic data spanning 11 crops over 11 fiscal years (2011/12–2021/22), the research evaluates Random Forest and Gradient Boosting Regressors across five distinct weather-feature datasets. The results demonstrate that the Random Forest Regressor consistently outperforms Gradient Boosting, explaining 74% to 84% of yield variance. The highest predictive accuracy was achieved using engineered growing-season averages (Baseline R2 = 0.83, Post-tuning R2 = 0.84), confirming that crop-specific seasonal summaries are the most informative features. Temperature emerged as the strongest single predictor of crop yield (R^2 = 0.82), while standalone precipitation exhibited the weakest predictive power (R2 = 0.74). These findings offer an empirical decision-support foundation for farmers and policymakers to quantify climate impacts and enhance sustainable agricultural planning in Nepal.

Keywords: predictive analytics, machine learning, crop yield prediction, Nepalese agriculture, weather impact

pages: 11-11, online: 2026