Table 2 Results for five proposed architectures across single categories using two preprocessing methods: Full-band and TD-F under random split evaluation strategy.

From: EEG imagined speech neuro-signal preprocessing and deep learning classification

Architecture

Category

Characters

Digits

Full-band

TD-F

vs. Baseline (CNN-1-LSTM)

Full-bands

TD-F

vs. Baseline (CNN-1-LSTM)

CNN-1-LSTM (baseline)

Accuracy(%)

87.00 ± 0.40

87.00 ± 0.40

–

87.00 ± 0.40

87.65 ± 0.40

–

Precision(%)

87.00 ± 0.40

87.82 ± 0.40

–

87.00 ± 0.40

87.88 ± 0.40

–

CNN-2-LSTM

Accuracy(%)

92.63 ± 0.26

92.92 ± 0.26

+ 5.92

93.16 ± 0.26

94.11 ± 0.26

+ 6.46

Precision(%)

92.65 ± 0.26

92.97 ± 0.26

+ 5.15

93.19 ± 0.26

94.14 ± 0.26

+ 6.26

CNN-2-Bi-LSTM

Accuracy(%)

92.11 ± 0.22

93.85 ± 0.22

+ 6.85

92.92 ± 0.22

92.70 ± 0.22

+ 5.92

Precision(%)

92.17 ± 0.22

93.89 ± 0.22

+ 6.07

92.97 ± 0.22

92.74 ± 0.22

+ 5.97

CNN-3-LSTM

Accuracy(%)

91.43 ± 0.28

93.16 ± 0.28

+ 6.16

91.97 ± 0.28

93.09 ± 0.28

+ 5.44

Precision(%)

91.50 ± 0.28

93.16 ± 0.28

+ 5.34

92.03 ± 0.28

93.14 ± 0.28

+ 5.25

3-LSTM

Accuracy(%)

88.14 ± 0.39

90.85 ± 0.39

+ 3.85

88.81 ± 0.39

92.32 ± 0.39

+ 4.67

Precision(%)

88.18 ± 0.39

90.91 ± 0.39

+ 3.09

88.87 ± 0.39

92.39 ± 0.39

+ 4.51

 

Objects

   

Full-bands

TD-F

vs. Baseline (CNN-1-LSTM)

CNN-1-LSTM (baseline)

Accuracy(%)

86.00 ± 0.40

86.00 ± 0.40

–

   

Precision(%)

86.19 ± 0.40

86.00 ± 0.40

–

   

CNN-2-LSTM

Accuracy(%)

91.65 ± 0.26

93.78 ± 0.26

+ 7.78

   

Precision(%)

91.67 ± 0.26

93.82 ± 0.26

+ 7.82

   

CNN-2-Bi-LSTM

Accuracy(%)

93.27 ± 0.22

92.16 ± 0.22

+ 7.27

   

Precision(%)

93.30 ± 0.22

92.19 ± 0.22

+ 7.11

   

CNN-3-LSTM

Accuracy(%)

90.31 ± 0.28

90.38 ± 0.28

+ 4.38

   

Precision(%)

90.35 ± 0.28

90.44 ± 0.28

+ 4.44

   

3-LSTM

Accuracy(%)

88.56 ± 0.39

89.77 ± 0.39

+ 3.77

   

Precision(%)

88.59 ± 0.39

89.77 ± 0.39

+ 3.77

   
Back to article page