Smart Crop Selection and Yield Prediction Through Machine Learning and Explainable Artificial Intelligence

Abstract

Effective agricultural decision making requires both the identification of a suitable crop under prevailing soil and climatic conditions and the estimation of its expected productivity. This study proposes an integrated dual model machine learning framework that combines crop recommendation with crop yield prediction. The crop recommendation component was developed using 2,200 observations representing 22 crop classes and seven soil and climatic attributes, namely nitrogen, phosphorus, potassium, temperature, humidity, soil pH, and rainfall. The yield prediction component utilized a historical crop production dataset containing 246,091 records, which was reduced to 240,505 observations after preprocessing. The processed dataset covered 90 crops, 33 states, 646 districts, and agricultural records from 1997 to 2015. Decision Tree, Gaussian Naive Bayes, Support Vector Machine, Logistic Regression, and Random Forest were evaluated for crop recommendation, while Linear Regression, Decision Tree Regression, and Random Forest Regression were compared for yield estimation. The selected Random Forest classifier achieved an accuracy of 99.55 percent, precision of 99.57 percent, recall of 99.55 percent, and macro F1 score of 99.55 percent. For crop yield prediction, Random Forest achieved a coefficient of determination of 0.9475, a mean absolute error of 0.7033, and a root mean squared error of 2.6432. Statistical analysis confirmed significant overall performance differences among the evaluated classification and regression models. Explainability analysis further identified humidity as the most influential variable for crop recommendation and season as the dominant contributor to yield prediction. The proposed framework integrates crop selection, crop mapping, yield estimation, production estimation, statistical validation, and interpretable machine learning within a unified agricultural decision support system.