Sustainable Paddy Fertilization Strategy Using Ensemble Learning Techniques and Soil Test Crop Response Integration in Chhattisgarh Plains

Abstract

Fertilizer management in paddy cultivation requires precise nutrient recommendations to optimize productivity while minimizing environmental impacts and production costs. This study presents an intelligent fertilizer recommendation system that integrates ensemble machine learning techniques with Soil Test Crop Response (STCR) methodology for sustainable paddy production in Chhattisgarh Plains. The system combines Support Vector Regression, Random Forest, and Gradient Boosting algorithms through a stacking ensemble approach to predict optimal Nitrogen (N), Phosphorus (P), and Potassium (K) requirements based on soil test results and target yield objectives. Data from 847 farmer fields across 15 districts in Chhattisgarh Plains were analyzed, incorporating comprehensive soil health parameters and historical fertilizer response data. The ensemble system achieved superior accuracy in nutrient recommendations compared to conventional STCR calculations and individual machine learning models, with mean absolute errors of 8.2 kg/ha for nitrogen, 3.6 kg/ha for phosphorus, and 7.8 kg/ha for potassium predictions. Economic analysis revealed potential cost savings of 18-24% in fertilizer expenditure while maintaining yield targets, supporting sustainable intensification objectives in rice production systems.