SIMLR: Machine Learning inside the SIR Model for COVID-19 Forecasting
Problem
Forecasting the number of newly infected people one to four weeks ahead during the COVID-19 epidemic is hard because government policies change over time.
Why it matters
Accurate infection forecasts enable timely, effective public-health decisions.
What the AI did
SIMLR embeds machine learning in the SIR epidemiological model, tracking government policy changes to estimate time-varying parameters and forecast new infections plus probabilities of future policy changes.
Results
Mean average percentage error as good as state-of-the-art forecasting models on Canada and US data, with the added advantage of interpretability (no exact numbers in abstract).
Impact
Applied to Canadian and US data with competitive accuracy, but no exact error values are given.
Record
| Venue | Forecasting |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Roberto Vega, Leonardo Albitres Flores, Russell Greiner |
| Citations | 38 (OpenAlex, at import time) |
| References | 36 |
| Model or method | SIMLR (ML-augmented SIR model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/forecast4010005 |
| Open access | gold |
| Keywords | COVID-19 forecasting, SIR model, machine learning, policy tracking, interpretable model |
Importance rating
major — millions of lives, an industry or a field
Addresses forecasting during the COVID-19 epidemic, a pandemic-scale problem; abstract stresses accurate forecasts are critical for timely public-health decisions and applicability to other infectious diseases.