Abstract

ABSTRACT
Aim/Background: Jhum (shifting) cultivation remains a vital agricultural practice in the Chittagong Hill Tracts of Bangladesh but faces challenges related to soil degradation, climate variability, and limited data-driven decision support. This study aims to demonstrate a robust machine learning–based crop recommendation framework as a proof-of-concept for sustainable crop selection.
Methods: A publicly available crop recommendation dataset comprising 2,201 records with soil nutrients (N, P, K), temperature, humidity, pH, and rainfall was used. An ensemble learning framework based on soft voting was developed using Random Forest, Decision Tree, and Support Vector Machine classifiers. Stratified 10-fold cross-validation was applied, and performance was evaluated using accuracy, precision, recall, F1-score, and RMSE. Explainable AI was incorporated using SHAP to interpret model predictions.
Results: The proposed ensemble model achieved a mean cross-validated accuracy of 99.14%, demonstrating stable and consistent performance across folds. SHAP analysis indicated that nitrogen, rainfall, and potassium were among the most influential features driving crop recommendations.

Conclusion: Although the dataset does not explicitly represent jhum-specific conditions, the study demonstrates the potential of ensemble machine learning combined with explainable AI for transparent and robust crop recommendation. The framework provides a methodological foundation for future region-specific and real-time decision-support systems aligned with sustainable agriculture goals.

Key words: Shifting Cultivation, Sustainable Agriculture, Crop Recommendation, Ensemble Machine Learning, Explainable AI