Table 2 Statistical test results of model performance.

From: Dynamic forecasting of China’s carbon market prices by the coupling of macroeconomic indicators and LSTM model

Model type

Mean absolute error (MAE)

Prediction accuracy (%)

MAE difference from LSTM-CNN

Wilcoxon p-value

Significance

LSTM-CNN

0.33

89.2

LSTM only

0.52

78.5

0.19

0.002

**

CNN only

0.41

82.1

0.08

0.015

*

BiLSTM

0.47

80.3

0.14

0.005

**

GRU

0.50

79.1

0.17

0.003

**

Transformer

0.61

75.8

0.28

0.001

**

GAN

0.58

76.5

0.25

0.001

**