DeepCOVIDNet: An Interpretable Deep Learning Model for Predictive Surveillance of COVID-19 Using Heterogeneous Features and Their Interactions.

Ankit Ramchandani,Ali Mostafavi,Chao Fan

doi:10.1109/access.2020.3019989

Ankit Ramchandani, Ali Mostafavi + Show 1 more

Open Access

https://doi.org/10.1109/access.2020.3019989

Copy DOI

Abstract

In this paper, we propose a deep learning model to forecast the range of increase in COVID-19 infected cases in future days and we present a novel method to compute equidimensional representations of multivariate time series and multivariate spatial time series data. Using this novel method, the proposed model can both take in a large number of heterogeneous features, such as census data, intra-county mobility, inter-county mobility, social distancing data, past growth of infection, among others, and learn complex interactions between these features. Using data collected from various sources, we estimate the range of increase in infected cases seven days into the future for all U.S. counties. In addition, we use the model to identify the most influential features for prediction of the growth of infection. We also analyze pairs of features and estimate the amount of observed second-order interaction between them. Experiments show that the proposed model obtains satisfactory predictive performance and fairly interpretable feature analysis results; hence, the proposed model could complement the standard epidemiological models for national-level surveillance of pandemics, such as COVID-19. The results and findings obtained from the deep learning model could potentially inform policymakers and researchers in devising effective mitigation and response strategies. To fast-track further development and experimentation, the code used to implement the proposed model has been made fully open source.

Highlights

COVID-19 has had an unprecedented social and economic impact worldwide
Other forecasting models proposed for other epidemics [9]–[11] rely on features like human mobility and within-season and between-season observations
To complement existing epidemiological models, we propose a deep learning model based on the high-level framework of DeepFM [22] that takes in multiple features, accounts for interactions between them and forecasts growth in the number of infected cases in all U.S counties

Summary

INTRODUCTION

COVID-19 has had an unprecedented social and economic impact worldwide. With more than 13 million infected cases and more than half a million deaths as of mid-July, the pandemic is still accelerating globally without showing any signs of nearing an end. Visitation patterns capture the number of visits to important types of POIs, Venables distance captures the agglomeration of population activities in a county, and social distancing metrics show adherence to stay-at-home orders Together, these three feature groups show the extent to which people are likely to come in close contact and potentially facilitate the spread of infection. DeepCOVIDNet We have designed a novel deep learning model, DeepCOVIDNet, to estimate the range of increase in the number of infected cases on a particular day given multiple constant, time-dependent, and cross-county time-dependent feature groups as input. Due to this explicit representation of secondorder interactions, it is possible to identify pairs of features between which high amount of interaction is observed To perform this analysis, we evaluated the network on a section of the training set and tracked the magnitude of activations in the vector representing the second-order interactions of input features.

RESULTS

LIMITATIONS AND FUTURE

VIII. DISCUSSION AND CONCLUDING

DISEASE SPREAD ATTRIBUTES