Skip to main content

Table 2 T-test p-values perfomed between approach pairs for the resemblance metrics

From: Comparative assessment of synthetic time series generation approaches in healthcare: leveraging patient metadata for accurate data synthesis

Model

Approach comparison

DLA AUROCa

Correlation Similaritya

Autocorrelation MAEa

\({\boldsymbol{F}_{\boldsymbol{\gamma}}}\)a

DGANT

A1-A2

 ~ 1.00

3.39 × 108

9.33 × 104

3.47 × 108

A2-A3

7.46 × 10–1

7.25 × 1012

1.10 × 104

1.80 × 106

A3-A1

7.90 × 10–1

5.83 × 106

2.04 × 10–1

1.26 × 109

WGAN-GPT

A1-A2

8.36 × 10–1

5.56 × 1011

 ~ 0.00

2.30 × 1011

A2-A3

4.44 × 10–1

2.22 × 1016

 ~ 0.00

9.11 × 1012

A3-A1

2.64 × 10–1

 ~ 0.00

 ~ 0.00

5.37 × 10–1

DGANM

A1-A2

4.20 × 1013

9.17 × 106

5.67 × 10–2

1.82 × 1010

A2-A3

4.19 × 104

5.50 × 10–2

 ~ 0.00

3.15 × 10–1

A3-A1

8.90 × 1014

3.34 × 107

 ~ 0.00

8.88 × 1016

WGAN-GPM

A1-A2

1.36 × 106

1.18 × 107

1.46 × 109

3.42 × 103

A2-A3

5.60 × 105

2.20 × 10–11

 ~ 0.00

7.31 × 10–1

A3-A1

1.01 × 109

 ~ 0.00

 ~ 0.00

9.92 × 1012

DGANP

A1-A2

1.00

1.379 × 10–1

4.001 × 1010

1.811 × 106

A2-A3

1.00

1.806 × 106

4.563 × 1010

3.659 × 103

A3-A1

1.00

2.114 × 105

4.911 × 1012

1.029 × 109

WGAN-GPP

A1-A2

1.00

3.718 × 103

3.044 × 1012

8.882 × 1016

A2-A3

1.00

8.677 × 10–1

 ~ 0.00

1.222 × 107

A3-A1

1.00

1.510 × 104

4.441 × 1016

4.996 × 1014

  1. aSignificant results for a \(SL=0.05\) are underlined and bold