Propagation trend deduction model of major health emergencies in megacities based on Attention-Bi-LSTM
Zaipeng Duan, Jiong Li, Junjie Zhu, Yeyue Wu · 2024
With the continuous outbreak of major health incidents in megacities, China 's epidemic prevention and control has entered a new stage, facing new situations and new tasks. In order to effectively grasp the trend of public health events in the new stage, this study introduces a deep learning model to build a public health event trend prediction model. Firstly, a multivariate data set that conforms to the trend of virus evolution in the new stage of epidemic prevention is established. Then, the data were preprocessed by principal component analysis, normalization and other methods. After that, a variety of prediction models are built based on deep learning frameworks such as long-term and shortterm memory neural networks to compare the prediction effects, and the attention mechanism is increased to improve the prediction effect of the model. Finally, MSE and R 2 score are used to evaluate the effect of the model. Studies have shown that conventional environmental factors such as temperature and weather cannot be the eigenvalues of the trend prediction model, and the number of infections at the previous time step can more effectively predict the trend of viral infection. The prediction effect of the 'Attention-Bi-LSTM' model is significantly better than other machine learning and deep learning algorithms. The MSE is 0.014, and the R2 score is 0.560. The predicted number of new infections basically fits the real value.