Table 3 HASDM modeling results using MC dropout and direct probability prediction.

From: Uncertainty quantification techniques for data-driven space weather modeling: thermospheric density application

Metric

Set

MC dropout (%)

Direct probability (%)

Error

Training

9.07

8.55

Validation

10.69

9.91

Test

10.69

10.60

Calibration

Training

3.06

1.74

Validation

2.51

2.45

Test

1.76

2.81

  1. Error refers to mean absolute error, and calibration is computed using Eq. (8).
  2. Bold values represent the lowest error between methods.
Back to article page