A hybrid ICA-GRU approach for forecasting gold futures price

Accurate forecasting of gold futures prices is very important for investors and analysts due to the highly volatile and nonlinear nature of the commodity markets. This research work develops a hybrid ICA-GRU framework to gold futures price forecasting by developing a hybrid approach of ICA and GRU-based neural networks. The historical daily gold futures price data spanning from 2015 to 2026 were collected from Yahoo Finance. ICA transforms correlated financial variables into statistically independent components, generating informative latent feature representations that may better capture the underlying structure of financial time-series data for subsequent sequence learning. The GRU-based neural network was then applied to identify the nonlinear pattern in the data. The performance of the proposed ICA-GRU model was measured using RMSE, MAE, MASE and R ², precision, recall, confusion matrix, and significance testing. The experimental results demonstrated that the RMSE was 39.36, MAE was 26.35, MASE was 2.61 and R ² was 0.99, which was considerably better than that of the traditional GRU model. According to the results of the Wilcoxon Signed-Rank Test, the difference between model performances was statistically significant (p < 0.001). The obtained results suggest that the developed ICA-GRU model can be considered a promising tool for gold futures price forecasting.