Table 4 Forecast accuracies by algorithm and clinic using all features as well as MAPE accuracies of models using only autoregressive features in parentheses. Ranks per clinic are based on MAPE and are also combined across both
From: Forecasting patient flows with pandemic induced concept drift using explainable machine learning
| Â | Clinic 1 | Clinic 2 | Combined | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Â | MAPE | Rank | RMSE | MAE | MAPE | Rank | RMSE | MAE | Rank |
kNN | 17.5 ± 8.8 (11.3) | 12.7 | 29.8 | 25.2 | 21.7 ± 14.7 (15.0) | 12.3 | 15.7 | 13.2 | 12.5 |
SVR (NU) | 17.7 ± 11.3 (11.2) | 12.4 | 29.7 | 25.3 | 20.4 ± 16.8 (14.5) | 11.1 | 14.6 | 12.4 | 11.7 |
Benchmark | 16.7 ± 13.5 (16.7) | 11.4 | 25.6 | 22.0 | 21.9 ± 21.1 (21.9) | 11.1 | 14.6 | 12.4 | 11.2 |
Naive | 16.3 ± 11.7 (16.3) | 11.0 | 25.7 | 22.3 | 19.1 ± 16.6 (19.1) | 10.3 | 13.5 | 11.4 | 10.6 |
Prophet | 10.1 ± 5.3 (11.8) | 6.9 | 16.7 | 14.2 | 15.7 ± 14.6 (16.6) | 7.9 | 10.9 | 9.2 | 7.4 |
Naive (Enhanced) | 10.7 ± 6.1 (10.7) | 7.6 | 17.7 | 14.9 | 14.1 ± 9.4 (14.1) | 7.0 | 10.1 | 8.4 | 7.3 |
ARIMA | 10.5 ± 5.9 (10.5) | 7.4 | 17.3 | 14.6 | 14.4 ± 9.6 (14.1) | 7.2 | 10.1 | 8.5 | 7.3 |
Kernel Ridge | 9.9 ± 4.2 (10.6) | 7.0 | 16.8 | 14.1 | 14.1 ± 9.0 (14.3) | 7.5 | 10.1 | 8.3 | 7.3 |
Ridge | 9.9 ± 4.2 (10.6) | 6.9 | 16.8 | 14.1 | 14.1 ± 9.2 (14.3) | 7.3 | 10.1 | 8.3 | 7.1 |
Random Forest | 10.0 ± 5.6 (10.5) | 7.1 | 16.7 | 13.9 | 13.9 ± 9.1 (14.2) | 7.1 | 10.0 | 8.3 | 7.1 |
CatBoost | 9.6 ± 4.4 (10.4) | 6.8 | 16.3 | 13.5 | 13.4 ± 7.0 (14.2) | 7.1 | 9.8 | 8.1 | 7.0 |
Gradient Boosting | 9.6 ± 4.5 (10.3) | 6.6 | 16.3 | 13.5 | 13.6 ± 7.6 (14.1) | 7.1 | 9.9 | 8.2 | 6.8 |
Stacking | 9.4 ± 4.8 (10.0) | 5.8 | 15.8 | 13.2 | 13.6 ± 10.4 (13.9) | 6.2 | 9.6 | 8.0 | 6.0 |
Averaging | 9.2 ± 4.6 (9.9) | 5.1 | 15.3 | 12.8 | 13.3 ± 9.4 (13.7) | 5.6 | 9.5 | 7.9 | 5.4 |
Voting | 9.0 ± 4.6 (9.7) | 5.0 | 15.0 | 12.6 | 13.1 ± 9.1 (13.5) | 5.4 | 9.3 | 7.8 | 5.2 |