Spatiotemporal disease case prediction using contrastive predictive coding
3rd ACM SIGSPATIAL International Workshop on Spatial Computing for Epidemiology, SpatialEpi 2022
; : 26-34, 2022.
Article
in English
| Scopus | ID: covidwho-2153137
ABSTRACT
Time series prediction models have played a vital role in guiding effective policymaking and response during the COVID-19 pandemic by predicting future cases and deaths at the country, state, and county levels. However, for emerging diseases, there is not sufficient historic data to fit traditional supervised prediction models. In addition, such models do not consider human mobility between regions. To mitigate the need for supervised models and to include human mobility data in the prediction, we propose Spatial Probabilistic Contrastive Predictive Coding (SP-CPC) which leverages Contrastive Predictive Coding (CPC), an unsupervised time-series representation learning approach. We augment CPC to incorporate a covariate mobility matrix into the loss function, representing the relative number of individuals traveling between each county on a given day. The proposal distribution learned by the algorithm is then sampled by the Metropolis-Hastings algorithm to give a final prediction of the number of COVID-19 cases. We find that the model applied to COVID-19 data can make accurate short-term predictions, more accurate than ARIMA and simple time-series extrapolation methods, one day into the future. However, for longer-term prediction windows of seven or more days into the future, we find that our predictions are not as competitive and require future research. © 2022 ACM.
Full text:
Available
Collection:
Databases of international organizations
Database:
Scopus
Type of study:
Prognostic study
Language:
English
Journal:
3rd ACM SIGSPATIAL International Workshop on Spatial Computing for Epidemiology, SpatialEpi 2022
Year:
2022
Document Type:
Article
Similar
MEDLINE
...
LILACS
LIS