Table 3b.
14-day forecast using time-series regression baselines, SSIR-HCD, and Neural Network SIR-HCD model. The 14-day forecast period was from September 13 to September 26, 2020.
| Method | Training period | Hospitalization MSLE | Mortality MSLE | Hospitalization MAE | Mortality MAE | |
|---|---|---|---|---|---|---|
|
| ||||||
| Baselines | Simple Exponential Smoothing | Phase 1 and Phase 2 | 0.01169 | 0.00155 | 36.35 | 86.07 |
| Auto Regression | Phase 1 and Phase 2 | 0.01807 | 0.00001 | 44.99 | 5.05 | |
| ARIMA | Phase 1 and Phase 2 | 0.10973 | 0.00007 | 89.50 | 18.38 | |
| Double Exponential Model | Phase 1 and Phase 2 | N/A | 0.58694 | N/A | 1320.90 | |
| YYG | Phase 1 and Phase 2 | N/A | 0.00003 | N/A | 11.04 | |
| Vanilla SIR-HCD | Phase 1 and Phase 2 | 0.03671 | 0.00037 | 59.49 | 47.17 | |
| Proposed models | SSIR-HCD with step function | Phase 1 and Phase 2 | 0.01444 | 0.00009 | 36.53 | 22.79 |
| SSIR-HCD with exponential function | Phase 1 and Phase 2 | 0.00012 | 0.00180 | 58.25 | 26.44 | |
| Nonlinear extension of SIR-HCD* with neural networks | Phase 1 and Phase 2 | 0.03984 | 0.77674 | 453.30 | 224.77 | |
*
See Appendix for more details.