Table 3 Surrogate model performance evaluation using R-squared statistic.

From: Designing accurate emulators for scientific processes using calibration-driven deep models

Test case

Methods

 

RF

GBT

DNN

DNN (drp)

LbC

Grid stability

0.89 ± 0.008

0.85 ± 0.007

0.94 ± 0.006

0.96 ± 0.003

0.97 ± 0.002

Concrete

0.84 ± 0.22

0.82 ± 0.21

0.88 ± 0.13

0.89 ± 0.14

0.91 ± 0.09

Parkinsons

0.71 ± 0.12

0.69 ± 0.14

0.7 ± 0.11

0.71 ± 0.13

0.75 ± 0.11

Superconductivity

0.84 ± 0.17

0.79 ± 0.15

0.84 ± 0.19

0.86 ± 0.21

0.89 ± 0.13

Airfoil self-noise

0.89 ± 0.11

0.81 ± 0.19

0.88 ± 0.12

0.9 ± 0.11

0.94 ± 0.06

ICF JAG (scalars)

0.995 ± 0.002

0.983 ± 0.003

0.975 ± 0.005

0.991 ± 0.002

0.998 ± 0.001

ICF Hydra (scalars)

0.88 ± 0.015

0.81 ± 0.019

0.88 ± 0.08

0.89 ± 0.09

0.94 ± 0.08

ICF Hydra (multi)

0.87 ± 0.011

0.81 ± 0.03

0.91 ± 0.01

0.95 ± 0.006

0.97 ± 0.008

Reservoir

0.89 ± 0.004

0.87 ± 0.008

0.91 ± 0.01

0.93 ± 0.005

0.96 ± 0.006

  1. The results were obtained over fivefold cross-validation, carried out using three different random seeds, on each of the use cases using emulators designed with different machine-learning approaches. We report the mean and standard deviation across different trials, and the best performance in each case is denoted in bold.
Back to article page