Hybrid Rule-Based and Machine Learning Decision-Support Framework for Smart Paddy Farming

Abstract

Paddy farming decisions are affected by changing weather conditions, water availability, soil conditions, and crop health, creating a need for decision-support approaches that combine agronomic knowledge with data-driven prediction. This research presents a hybrid rule-based and machine-learning decision-support solution for paddy management that combines interpretable agronomic rules and lightweight predictive models. The temporal alignment, interpolation, normalization, feature engineering, and crop-stage aggregation of weather and NDVI data were integrated and preprocessed. It is a rule-based engine with 63 expert rules that was fused with Decision Tree, Logistic Regression, and Random Forest models using a weighted decision-fusion mechanism. The hybrid model performed with 90% accuracy, 0.91 precision, 0.88 recall, and 0.89 F1-score, while the rule-based model had 72% accuracy, and the best standalone Random Forest model had 86% accuracy. The hybrid model also yielded the minimum RMSE of 7.6. The results of the simulation scenarios showed increases in yield of 3–17% and savings in irrigation-water use of 12.4%. There were 10 farmers and 2 agricultural officers who participated in the preliminary usability test and were scored from 4.0 to 4.8 out of 5. The key is an adaptive machine-learning prediction integrated with clear agronomic logic, in a lightweight advisory system. The yields and water-saving benefits, however, are simulated, and experimental testing on a wider range of locations and seasons will be needed to determine the full agronomic benefits.