Machine Learning for Crop Recommendation and Yield Forecasting: A Practical Framework for Sustainable Agriculture

Abstract

Agriculture is increasingly influenced by changing weather, soil conditions, water availability, and input costs. Farmers often have to make crop-selection and production decisions with incomplete or delayed information. This paper presents a practical machine-learning framework that connects basic soil and weather observations with crop recommendation and yield forecasting. The proposed workflow includes data preparation, feature selection, supervised learning, model validation, and explainability. For crop recommendation, we discuss classification methods such as Random Forest, Extra Trees, Logistic Regression, Decision Tree, K-Nearest Neighbour, Gaussian Naive Bayes, and Support Vector Machine. In contrast, regression methods such as Decision Tree Regressor, Random Forest Regressor, KNN Regressor, Support Vector Regressor, and XGBoost Regressor are considered for yield estimation. The supplied reference study reported very high benchmark performance on its datasets, including 99.7% classification accuracy for Random Forest and an R² value of 99.99% for SVR. We treat these numbers as reported benchmark results rather than new experimental results. The main contribution of this rewritten paper is a clearer, application-oriented framework that explains how machine learning can support agricultural decisions while recognizing data quality, geographic transferability, interpretability, and real-world deployment limitations