Optimized Ensemble Learning for Enhanced Crop Recommendation: Leveraging ML for Smarter Agricultural Decision-Making

Agriculture is a country's backbone and plays an essential role in shaping its economic performance. Factors like disasters, extreme weather changes, pests, and soil quality greatly impact’s productivity resulting in economic loss. Accurate predictions in agricultural practices, such as crop recommendations, can significantly enhance productivity and resource management. The objective of this research is to build a robust crop recommendation system using an ensemble model composed of several regression machine learning (ML) models. The study uses a real-time dataset collected using IoT sensors for crop recommendation of 22 different crop varieties. The dataset is available in Kaggle. The main features of the dataset are nitrogen, phosphorus, potassium and Ph value of soil, humidity and rainfall. This study compares the performance of regression models with ensemble models and the impact of different hyperparameter tuning techniques such as Bayesian Optimization, Genetic algorithm and GridSearchCV. Fune-tuning is done to improve the predictive performance thus providing smarter agriculture. Regression Models like Logistic Regression, SVM, Decision Tree, Navie Bayes, K Nearest Neighbour, Extra Tree Classifier, XGBoost, Gradient Boost models are compared with ensemble models like voting, bagging, boosting and stacking ensemble model. Among all the ensemble techniques stacking ensemble obtained a highest accuracy of 99.3% when compared to other regression models.