IMPLEMENTATION OF GATED RECURRENT UNIT (GRU) WITH PARTICLE SWARM OPTIMIZATION (PSO) FOR PALM OIL PRICE PREDICTION

Abstrak

Palm oil is a major agricultural commodity whose price fluctuates over time, creating uncertainty for planning and decision-making. This study develops a time-series prediction model for Palm Oil Futures prices by implementing a Gated Recurrent Unit (GRU) and optimizing selected GRU hyperparameters using Particle Swarm Optimization (PSO). The dataset was obtained from Investing.com and contains 2,376 daily observations covering 2 June 2016 to 2 June 2026, with Date and Price used as the principal attributes. The research workflow includes data cleaning, chronological train-test splitting, Min-Max normalization, sequence construction, baseline GRU development, PSO-based hyperparameter optimization, final model training, and evaluation using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). Three split scenarios (70:30, 80:20, and 90:10) and two sequence lengths (14 and 28) were examined before the final configuration was selected. PSO searched GRU Units and Dense Units in the ranges 64-128 and 32-64, respectively. The best baseline configuration achieved RMSE 2.890, MAE 2.303, and MAPE 6.681%, while the optimized GRU-PSO achieved RMSE 1.956, MAE 1.653, and MAPE 6.689%. Thus, PSO reduced RMSE by about 32.3% and MAE by about 28.2%, while MAPE increased slightly. The final model was also used to generate a seven-day short-term forecast.