Multi-Agent System Based on Stacking Technique for Rice Yield Prediction

Artificial intelligence (AI) has expanded its influence across various sectors including education, healthcare, and agriculture. In the agricultural setting, the multiagent system (MAS) is recognized as a powerful tool for optimizing resource allocation, enhancing decision-making processes, and improving overall farm productivity. Accurate prediction of rice yield levels is paramount importance in the agricultural sector. It allows farmers, policymakers, and stakeholders to make informed decisions regarding crop management. Individual machine-learning algorithms (MLA) have been used to predict rice yield levels, but they may not fully exploit the available information. Therefore, the novel system implemented based on stacking techniques to solve complex problems. The objective of this paper is to present an efficient system that leverages MAS based on a novel proposed stacking technique (Extra Trees Classifier (ETC), Random Forest Classifier (RFC), Linear Discriminant Analysis (LDA) and Gaussian Naive Bayes (GNB)) for improving the rice yield level prediction within the agricultural environment. Each algorithm brings its unique approach to handling complex relationships, and modeling class separability. The findings from this study provided valuable insights for decision-making in interconnected sectors and facilitating optimal business planning. The dataset incorporated climate variables such as monthly maximum and minimum temperatures and rainfall. The final result of the dataset consists of 1266 rows and 18 features. The results showed that the novel proposed stacking technique achieved the highest prediction accuracy 87% and the best individual Decision tree classifier obtained 77.5%. The novel proposed stacking technique increases the accuracy by 9.5% compared to the best individual MLA.