Advanced bagging ensemble technique for multi-crop predictive modeling to enhance agricultural decision-making

Abstract

Crop production is a cornerstone of agriculture, significantly influencing economies and farmers' livelihoods. However, fluctuating environmental conditions complicate the selection of suitable crops, requiring expertise in factors such as soil type, climate, humidity, rainfall, and temperature. Existing crop recommendation models primarily focus on a limited range of crops, such as rice, maize, and wheat, which restricts their utility across varied agricultural settings. Additionally, these models often exhibit inconsistent accuracy and high false-positive rates, undermining their reliability for practical use. To overcome these challenges, this study proposes a Bagging-based ensemble model that integrates seven machine-learning algorithms: Decision Tree, Support Vector Machine, Logistic Regression, Naive Bayes, Random Forest, K-Nearest Neighbor, and XGBoost. Leveraging a dataset enriched with diverse environmental and soil features—using soil type encoding and feature normalization—the model captures complex relationships that influence crop suitability. The ensemble model demonstrates an outstanding 99.9% accuracy, with macro-average precision, recall, and F1 scores of 99%, surpassing traditional models in performance. This advanced predictive tool offers a robust and versatile solution, enabling accurate and adaptable crop recommendations to support farmers and agricultural stakeholders in diverse environmental conditions.