EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
3431. Forecasting high dimensional Tensor with relatively few observations.
Invited abstract in session TC-40: Data mining biomedical applications, stream Advances in Stochastic Modelling and Learning Methods.
Tuesday, 12:30-14:00Room: 96 (building: 306)
Authors (first author is the speaker)
| 1. | Nuria Diaz-Tena
|
| MSIS, Rutgers University |
Abstract
When studying excess Covid deaths in the 50 US states, the responses is a tensor time series of 50 states by 15 major causes of deaths and the overall. This data is observed monthly since 2015 before covid until 2023 and the objective is to estimate the excess deaths during covid years.
We propose a model that combines a non-linear trend model that explains the over-all structure of the data, followed by a seasonality model, and a tensor autoregressive model for the residuals.
The seasonality model consists in a combination of sinusoidal trends at different frequencies.
The results give clear patters of excess deaths by state and disease during covid time that can be communicated and interpreted with data visualization.
Keywords
- Analytics and Data Science
- Computational Biology, Bioinformatics and Medicine
Status: accepted
Back to the list of papers