Combining predictive models: Hybrid approaches in crop recommendation systems

Abstract

This study aims to develop a robust crop recommendation system by using various machine learning (ML) algorithms and hybrid models. The dataset, sourced from Kaggle, having essential agricultural parameters, including nitrogen, phosphorus, potassium, soil pH, humidity, temperature and rainfall, across 22 crops. Using algorithms such as Decision Trees, Random Forest, Naïve Bayes, Support Vector Machines, Logistic Regression and XGBoost, this research evaluates both standalone and hybrid models to determine the optimal approach for accurate crop recommendations. Results indicate that individual models like Random Forest and Gaussian Naïve Bayes perform exceptionally well, achieving accuracies close to 99%, while hybrid models, particularly the SVM + GNB combination, also demonstrate equally strong predictive performance. These findings suggest that a carefully chosen individual model may be equally effective, though hybrid models offer further refinement in certain scenarios. The proposed system provides farmers with data-driven insights for crop selection based on soil and environmental conditions, supporting precision agriculture and sustainable farming practices.