Nonlinear Economic Forecasting Model under Artificial Neural Network

In response to the problem of difficulty in capturing nonlinear relationships in economic forecasting and poor prediction accuracy, this article takes GDP and growth rate as research objects and uses Long Short-Term Memory (LSTM) model to study nonlinear economic forecasting. Firstly, this article uses principal component analysis to select features from the self-built dataset, which speeds up the training time. Then, the LSTM model is used to focus on the key feature points of the predicted data sequence in a targeted manner. It better captures long-distance dependency and nonlinear relationship changes in the sequence, reducing the error of economic prediction. Finally, this article compares the predictive capabilities of five models, LSTM, Gated Recurrent Unit (GRU), Support Vector Machine (SVM), Decision tree, and Random forest, for nonlinear relationships. The experimental results show that the LSTM model has an average absolute percentage error (MAPE) of 1.2% in predicting gross domestic product. The Mean Absolute Error (MAE) reached 120.4 trillion yuan, reducing the prediction error of non-linear economic relationships.