Optimal Forecasting for Iron Ore Price Fluctuations With Hybrid Residual/Shuffle Networks and Enhanced Spring Search Optimization

ABSTRACT Anticipating the iron ore cost volatility is also valuable information for the stakeholders in the mining and steel industries, which would also allow for minimizing the risk and maximizing the benefit of the decision‐making process. Here, we proposed a hybrid framework with an ESSA (Enhanced Spring Search Algorithm) to incorporate Shuffle/Residual Networks to achieve a more precise prediction. This is precisely where the Residual/Shuffle Network uses the ability of deep learning architectures to capture the information retained in complex non‐linear price time series data, which are represented by the historical prices of iron ore. At the same time, the ESSA attracts the best hyperparameter solutions and upgrades model efficiency using a prominent metaheuristic. The findings indicate that the proposed ESSA‐RSN model is significantly superior to the state‐of‐the‐art methods, including Long Short‐Term Memory (LSTM) network, Extreme Learning Machine (ELM), Support Vector Regression (SVR), Random Forest Regressor (RFR), and Gradient Boosting Regressor (GBR), producing improvement values of up to 30.64% when measured via MSE, 14.55% via RMSE, and 6.35% when gauged against MAE with regard to LSTM for the 80:20 split. For 90:10, it showed superior performance with an improvement of 21.65%, 8.58%, and 4.58%, thereby establishing its robustness even under limited training data. This shows the superiority of the proposed framework based on different performance metrics and confirms its robustness relative to both classical machine learning and modern deep learning approaches. These encouraging results show the model's effectiveness in capturing the nuances of the volatile, non‐linear commodity markets and hence prove to be a sound tool for forecasting in practice.