SPOTCPC: A Novel Approach to COVID-19 Prediction Utilizing Human Mobility Data and Contrastive Predictive Coding

Overview

During the COVID-19 pandemic, time series prediction models have been essential in informing policymaking and response efforts by forecasting cases and deaths from the country to county levels. However, the emergence of new COVID-19 variants presents challenges for existing models, which may not account for previously unseen disease trends. Furthermore, several models fail to incorporate covariate features, such as human mobility data, which could enhance model accuracy due to their correlation with COVID-19 spread. T o address these challenges, we propose SPOTCPC (Spatial PrObabilisTic Contrastive Predictive Coding), which augments CPC by incorporating a mobility matrix that represents the relative number of individuals traveling between each county on a given day into the model’s loss function. The Metropolis-Hastings algorithm then samples the proposal distribution learned by this component of SPOTCPC to provide a final prediction of the number of COVID-19 cases in each region. Our experiments show that SPOTCPC can make accurate short-term predictions, which are more accurate than ARIMA and other time-series extrapolation methods, one day into the future. We also find that the SPOTCPC outputs for prediction windows seven or more days into the future are comparable to those provided by existing models.

Competition history

• JSHS 2023 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.