Machine Learning-Based Multi-Crop Soil Suitability Classification for the Bogura Region of Bangladesh

This study presents a region-specific machine learning framework for multi-crop soil suitability classification in Bogura, Bangladesh. It integrates 1,698 geographic soil records with 11 soil and environmental parameters to independently evaluate the suitability of rice, maize, banana, and papaya. Four supervised machine learning models are compared, with XGBoost achieving the highest agreement with the rule-derived suitability classes. The study also incorporates feature-group ablation, class-wise evaluation, and ranked decision-rule extraction to improve model interpretation and support data-driven agricultural decision-making.