Application of a Hybrid Model Based on ICEEMDAN, Bayesian Hyperparameter Optimization GRU and the ARIMA in Nonferrous Metal Price Prediction

Abstract

Nonferrous metals are the basic materials for national economic development. Accurate and robust price forecasting can effectively reduce investment costs and bring greater economic benefits to enterprises. But the violent fluctuation of nonferrous metal prices and the evolution of irregular cycles make metal price prediction difficult. A nonferrous metal price prediction model based on improved complementary ensemble empirical mode (ICEEMDAN) decomposition, a Bayesian hyperparameter optimization gated recurrent neural network (GRU) and an integrated autoregressive moving average model (ARIMA) is proposed. First, ICEEMDAN is used to decompose the original metal price series, and then fuzzy entropy is applied to judge the complexity of confusion of the decomposed subsequence. An ARIMA model is used to predict the sequence with the minimum fuzzy entropy, and the other subsequences are predicted by the Bayesian hyperparametric optimization GRU in this article. Finally, the prediction results are integrated to obtain the final prediction results of the original price series. To verify the reliability and practicability of the model in this article, different lengths of daily futures price series of zinc, aluminum, copper and gold prices on the London Metal Exchange and the Investing are used as the research data, three experiments utilizing 5 groups of metal price data and 10 comparison models are designed to verify the superiority of the proposed model. The results show that: (i) Bayesian hyperparameter optimization can improve the prediction performance of a single model effectively. The hybrid method with fuzzy entropy can effectively improve the prediction accuracy. (ii) The proposed model outperforms the compared state-of-the art models in three-step metal prices forecasting.