A Medium to Long-Term Multi-Influencing Factor Copper Price Prediction Method Based on CNN-LSTM

Non-ferrous copper prices exhibit high noise, non-smoothness, and non-linearity, which pose significant challenges to accurate price prediction. One of the current methods for predicting copper prices is multi-influencing factor analysis, which typically relies on traditional optimization or neural network methods to identify factors that affect copper prices. However, extracting attribute features and high-level semantics from raw data using these conventional methods can be difficult, which may necessitate revision of the selected influencing factors and final results. This paper proposes a CNN-LSTM-based approach that leverages the feature extraction capabilities of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. After analyzing the fluctuation features of copper prices and their qualitative relationships with factors such as supply and demand, energy costs, alternative metals, global macroeconomic conditions, and national policies, we selected 11 influencing factors for copper price fluctuation as explanatory variables using scatter plots, Pearson correlation coefficients, and heat maps. These variables are then fed into a CNN-LSTM network as a two-dimensional multivariate time series, along with historical copper price data, for monthly price forecasting. Experimental results show that our proposed method outperforms other existing methods by utilizing the attribute space feature extraction capability of CNNs and the temporal feature extraction of LSTMs.