Economic monitoring and early warning based on feature screening and hybrid neural network

Intending to the issue that the existing study do not fully exploit features, the random forest algorithm (IRF) is improved first. The splitting feature screening function is simplified based on the principle of infinitesimal equivalence, and the Gini coefficient value of the non-category attribute is introduced to improve the computational efficiency of the algorithm. Then, public health economic impact variables are selected, and spatial features are extracted using a residual convolutional neural network. Temporal features are extracted using a gate rate unit (GRU), and a self-attention mechanism is incorporated to enhance the spatial and temporal features. Finally, the IRF filter is used to select the most important spatio-temporal features of the early warning results and map them to the monitoring and early warning results through nonlinear transformation. The experimental outcome indicates that the accuracy of the proposed model has been improved by 5.07%-14.85%.