AGRONOMIC CROP-MATCH: ENSEMBLE LEARNING WITH AGRO - CLIMATIC FEATURES FOR OPTIMAL CROP SUITABILITY

Abstract

This study addresses the critical challenge of helping farmers select the most suitable crops for their land conditions. We developed an intelligent recommendation system that analyzes soil properties, weather patterns, and climate data using advanced machine learning techniques. Our approach combines four powerful algorithms—Random Forest, XGBoost, LightGBM, and CatBoost—through a sophisticated stacking method that leverages each model's unique strengths. We conducted thorough investigations into which environmental factors matter most, carefully tuned each algorithm's parameters, and rigorously evaluated their performance. Testing on publicly available agricultural datasets showed that Random Forest excelled individually with 95.03% accuracy, while our combined stacking approach achieved 93.86% accuracy with notably better stability and resistance to overfitting. These findings demonstrate that ensemble methods hold significant promise for precision agriculture applications and real-world farming scenarios.