Considering that the non-linear path of monthly time-series for the iron ore imported to China is under reciprocal influences of multiple factors, the process of data generation is not easily represented in a time-series model. Based on the decomposition–integration method, superiorities of empirical mode decomposition (EMD), non-linear autoregressive neural network (NARNN), and autoregressive integrated moving average (ARIMA) models are integrated to establish a combined model EMD-NARNN-ARIMA. The empirical results show that, compared with the NARNN or seasonal autoregressive integrated moving average (SARIMA) models, the proposed model is more suitable for predicting data pertaining to the import of iron ore to China. The prediction error of EMD-NARNN-ARIMA is significantly lower than that of NAR and SARIMA, and, more importantly, it does not increase the time-complexity. The predicted result attained through the proposed model reveals that the import of iron ore to China from January 2019 to December 2020 will gradually decrease, accompanied by reasonable seasonal fluctuations, which is consistent with the decreasing trend in the demand for iron and steel as a result of the adjustment of China’s current industrial structure.