predictive modeling of iron ore prices using deepar with natural gas index as a correlated factor

The steel industry relies heavily on accurate iron ore price forecasting for optimized procurement and production planning. this study develops a predictive model using the deepar architecture to forecast iron ore prices, emphasizing the natural gas index as a correlated factor. utilizing a dataset of monthly prices from 1977 to the present, the model incorporates world bank data for its foundational analysis. key preprocessing steps included data cleaning, feature scaling, and dataset division into training and testing sets. the deepar model was pre-trained on a relevant dataset and fine-tuned with our specific data over 20 epochs. evaluation on the test set revealed a mean absolute error (mae) of 3.1907, a mean squared error (mse) of 16.1723, and a root mean squared error (rmse) of 4.0214, indicating strong predictive performance. compared to traditional methods like arima, the deepar model's probabilistic nature and capacity to handle covariates demonstrate its robustness for financial forecasting. the findings suggest that the model can significantly enhance strategic decision-making in the steel industry, particularly for iranian steelmakers, by providing reliable insights into future iron ore price movements amidst global market fluctuations.