Explainable XGBoost model for crop recommendation in Mizoram using hybrid random forest and particle swarm optimization

The study was carried out during 2023–2024 using multi-location data across the three districts of Lawngtlai, Serchhip, and Champhai to support sustainable agriculture in Mizoram, North-east India, with the help of XGBoost-based crop recommendation system. A hybrid feature selection approach combining Random Forest (RF) and Particle Swarm Optimization (PSO) was proposed to identify key agronomic features. Class imbalance was addressed using Synthetic Minority Oversampling Technique (SMOTE), and model performance was evaluated using standard metrics such as accuracy, precision, recall, and F1-score. GridSearchCV was employed for hyperparameter optimization, with a 5-fold cross-validation applied to validate model performance during training. The XGBoost classifier trained on the hybrid RF + PSO-optimized features outperformed those trained on the full feature set and the RF top-8 features, owing to the combined benefits of SMOTE-based class balancing and PSO-driven optimal feature selection. The SHAP analysis for the four major crops rice, maize, moong, and potato revealed that nitrogen (N) and potassium (K) were the most influential factors shaping crop prediction outcomes, followed by phosphorus (P) and soil pH, while rainfall had the least influence due to Mizoram’s consistently high and evenly distributed precipitation across its cultivation zones. The proposed approach enhances both accuracy and interpretability, providing a reliable decision-support framework for crop selection tailored to Mizoram’s diverse agro-climatic conditions.