Prediction of palm oil production using hybrid decision tree based on fuzzy inference system Tsukamoto

Abstract

This research addresses the challenge of optimizing rule creation for palm oil production at PT Tapiana Nadenggan. It deals with the complexity of diverse agricultural variables, environmental factors, and the dynamic nature of palm oil production. The existing problem lies in the limitations of conventional decision tree models—J48, reduced error pruning (REP), and random—in capturing the nuanced relationships within the intricate palm oil production system. The study introduces hybrid decision tree models—specifically J48-REP, REP-Random, and Random-J48—to address this challenge via combination scenarios. This approach aims to refine and update the rule creation process, enabling the recognition of nuanced performance processes within the selected decision tree combinations. To comprehensively tackle this challenge and problem, the study employs Tsukamoto’s fuzzy inference system (FIS) for a sophisticated performance comparison. Despite the complexity, intriguing results emerge after the forecasting process, with the standalone J48 decision tree achieving 85.70% accuracy and the combined J48-REP excelling at 93.87%. This highlights the potential of decision tree combinations in overcoming the complexities inherent in forecasting palm oil production, contributing valuable insights for informed decision-making in the industry.