A Hybrid Machine Learning Model for Location-Specific Crop Recommendation Using Soil and Climate Parameters

Abstract

Abstract Accurate crop recommendation systems are essential for optimizing agricultural productivity and sustainability, yet existing approaches often fail to integrate diverse environmental factors and adapt to location-specific conditions. This study proposes a hybrid machine learning model that leverages soil and climate parameters through a threestage pipeline: Random Forest for feature selection, Extreme Gradient Boosting for robust prediction, and a lightweight Feed forward Neural Network for final decision-making. The model was evaluated on real-world datasets, demonstrating superior performance with an overall accuracy of 95.3%, precision of 95.1%, recall of 95.0%, F1-score of 95.1%, and a root mean squared error (RMSE) of 0.12. Ablation experiments reveal that excluding Random Forest feature selection reduces accuracy to 91.9%, omitting geospatial adaptation lowers accuracy to 92.6%, and replacing the neural network with logistic regression drops accuracy further to 89.2%. These results confirm the effectiveness of the hybrid architecture and the critical role of feature selection and geospatial adaptation in enhancing crop recommendation accuracy. This work presents a scalable and locationsensitive framework with significant potential to advance precision agricultur