A novel copper price forecasting ensemble method using adversarial interpretive structural model and sparrow search algorithm

Copper is a commodity whose price can accurately and swiftly reflect changes in the macroeconomy due to its widespread use in the national economy and its strong industrial and financial characteristics. Copper prices are influenced by supply and demand, national policies, laws, and regulations, and its volatility poses many risks to producers, processors, and consumers. This study selects ten highly correlated factors as input features for the prediction model. These factors include the total copper inventories on the LME and COMEX, the US Dow Jones Industrial Average, Brent crude oil price, gold, silver, and iron ore futures prices, as well as the US dollar index, the exchange rate between the United States and China (USD/CNY) and Australia (USD/AUD), and the Shanghai copper price. To create a fully ensemble forecasting model, a convolutional neural network (CNN) was selected as the second layer model, and a deep extreme learning machine (DELM) was selected as the first layer single model alongside the extreme gradient boosting (XGBoost) algorithm and long and short-term memory networks (LSTM). The hyperparameters of each single model were further optimized with the sparrow search algorithm (SSA) to achieve the best results for the ensemble prediction model. The test results indicate that, compared to the single and semi-ensemble models, the fully ensemble model achieves the best copper price prediction results, with a mean absolute percentage error (MAPE) of 3.59%, and offers a novel approach for predicting copper price volatility. In addition, the adversarial interpretive structural model (AISM) is introduced to analyze the selected factors, and the influence degree of different categories of factors on the final forecast results is discussed, in order to verify the overall influence of the selected copper price factors on the forecast results.