STACKING ENSEMBLE-BASED PREDICTIVE SYSTEM FOR CROP RECOMMENDATION

Abstract

Agricultural sustainability relies on crop production, but the task of choosing appropriate crops for certain places is difficult owing to the ever-changing environmental circumstances. Traditional approaches are often limited in scope, failing to adapt to diverse soil types and environmental parameters. This study introduces a novel prediction method that utilizes a machine-learning model with ensemble approaches to provide recommendations for crops. The system was developed using a Design Science Research (DSR) methodology. The proposed model incorporates a wide array of machine-learning techniques, including K-Nearest Neighbors, Decision Trees, Support Vector Machines, Naive Bayes, Logistic Regression, and Extreme Gradient Boosting. The integration utilizes the Random Forest meta-model. The model was trained and validated using a large dataset gathered from Kaggle, which consisted of a wide variety of crops and environmental characteristics. The model's performance was evaluated using metrics such as Accuracy, Recall, F1-Score, and Precision. It exhibited outstanding accuracy of 99.8%, along with superior recall, precision, and F1 scores, outperforming previous research by a significant margin. Furthermore, data flow diagrams illustrate the data processing flow within the system. The implementation was carried out using the Python programming language, with MongoDB employed for database development. The resulting proof-of-concept system demonstrates the practical applicability of the model by providing reliable crop recommendations based on environmental data. This research marks a substantial advancement in optimizing crop management strategies through advanced predictive modeling, offering a robust tool to aid farmers in making informed decisions, ultimately enhancing agricultural productivity and sustainability.