Research on deep neural network-driven macroeconomic forecasting in the context of economic transformation

Abstract

Accurate prediction of macroeconomic development trends has a significant role in decision-making and preventive signaling for regional governments, industries and even residents. Based on the characteristics of macroeconomic development, this paper identifies five dimensions as the initial variables of the study: the number of employed population, fixed asset investment, financial expenditure, national bank loans, and scientific, educational and cultural inputs. Considering the volatility, correlation and systematic characteristics of the macroeconomic system as a whole, this paper introduces the neural network method as a forecasting tool for its development trend, and puts forward the time series forecasting and regression forecasting method based on neural network. And the linear mapping method is used to map the actual values and forecast values to the (0,1) interval. Region W is selected as the research sample, and the macroeconomic development forecasting model is constructed by setting assessment variables based on its macroeconomic performance during a total of ten years from 2006 to 2015. The proposed model shows significant macroeconomic forecasting performance compared to traditional statistical forecasting methods, with its root mean square error and average absolute error remaining below 0.001.