A Deep Residual Network Integrating Spatial-temporal Properties to Predict Influenza Trends at an Intra-urban Scale

Abstract

Influenza is one of the most common causes of human illness and death; thus, accurate and timely predictions for influenza trends are critical tasks for public health. Many studies have attempted to conduct influenza prediction at or beyond the city scale; however, larger spatial scales are too coarse to help analyze influenza epidemics or allow offering precise interventions inside a city. Moreover, the existing prediction models often ignore the spatial correlations of influenza activity between neighbouring regions although such correlations are potentially helpful in influenza prediction. To address the above issues, this study proposes an influenza prediction model based on a deep residual network that predicts influenza trends by integrating the spatial-temporal properties of influenza at an intra-urban scale. Using a real dataset of influenza in Shenzhen City, China, we tested our prediction model on 10 districts within the city. Our results show that our proposed deep residual model outperforms four baseline models, including linear regression (LR), artificial neural network (ANN), long short-term memory (LSTM) and spatiotemporal LSTM (ST-LSTM) models, thus demonstrating the effectiveness of the proposed prediction model. To our best knowledge, although deep-learning-based approaches have been shown to be useful in many fields in recent years, there has been no attempt to apply such approaches to influenza prediction. Therefore, this study is an initial attempt to introduce a deep learning model into influenza prediction. The proposed deep residual network is able to incorporate the spatial correlations of influenza, and it has obvious potential for making influenza predictions at finer spatial scales within a city, which can offer critical support for preciser public health interventions.