Table 3.
Performance of the resulting TPOT pipelines when using thickness, volume, and mean curvature information from each specific site separately.
| Site # | Pipeline | MAE |
|---|---|---|
| 0 | 3 Lasso + RVR + Ridge + RF | 5.557 |
| 1 | Lasso + KNR | 4.101 |
| 2 | ElasticNet + Extra Trees + Ridge | 4.721 |
| 3 | Linear SVR + RF | 4.027 |
| 4 | 2 Extra Trees + Ridge | 2.05 |
| 5 | RF | 6.667 |
| 6 | 2 GPR | 5.940 |
| 7 | 2 ElasticNet | 5.638 |
| 8 | ElasticNet + RF | 3.938 |
| 9 | Lasso + RF + Extra Trees | 6.685 |
| 10 | KNR + DT + Ridge | 9.210 |
| 11 | RVR | 4.213 |
| 12 | DT + Ridge | 4.375 |
| 13 | 2 RF + DT + Ridge | 10.155 |
| 14 | Extra Trees + 2 DT + LR + Ridge | 10.849 |
| 15 | LR | 1.861 |
| 16 | RF + ElasticNet + DT | 2.220 |
The results are presented as the mean MAE for a 5-fold cross validation. Lasso, lasso model fit with least angle regression; RVR, relevance vector regressor; Ridge, linear least squares with l2 regularization; RF, random forest; KNR, K-neighbors regressor; DT, decision tree; GPR, gaussian process regressor; LR, linear regression.