Comparative Analysis of ANFIS and State‐ANFIS for Forecasting Cooking Oil Prices Based on Processed Palm Oil Yield (Crude Palm Oil)

The adaptive neuro‐fuzzy inference system (ANFIS) is widely employed in modeling intricate systems, especially in forecasting cooking oil prices. However, ANFIS confronts limitations stemming from backpropagation, prompting the exploration of alternatives like particle swarm optimization (PSO). Hybrid PSO‐ANFIS models exhibit enhanced forecasting accuracy, albeit at the expense of increased computational time. Nonetheless, both ANFIS and hybrid PSO‐ANFIS encounter challenges in handling dynamic relationships influenced by macroeconomic factors. To address these issues, the development of the State‐ANFIS (S‐ANFIS) method integrates regime‐switching models, enhancing its capability to manage dynamic relationships. Particularly effective in cooking oil price prediction, S‐ANFIS clarifies the impact of external variables and improves forecast accuracy and interpretability by combining ANFIS with state‐space models. Our analysis underscores S‐ANFIS’s superiority over ANFIS, particularly with Gaussian membership functions, as it reduces RMSE and MAPE values by half while requiring fewer nodes, thereby improving computational efficiency. Additionally, integrating key state variables like crude palm oil (CPO) prices, inflation rates, and the USD exchange rate enhances the reliability of the model. Overall, S‐ANFIS offers a more accurate, interpretable, and efficient approach to forecasting cooking oil prices, demonstrating superior predictive capabilities.