Table 1 Hyperparameter configuration for TCFormer components: Transformer encoder, MK-CNN front-end, and TCN head. These values are applied consistently across all subjects and datasets.

MK-CNN

# of Temporal filters (\(\:{F}_{1}\))

32

Transformer encoder

# of layers (\(\:N\))

2

Kernel size (\(\:{K}_{C}\))

20, 32, 64

# of query heads (\(\:H\))

4

Depth multiplier (D)

2

# of key-value groups (\(\:G\))

2

1st pooling size (P₁)

8

Dropout rate (\(\:{p}_{e}\))

0.4

2nd pooling size (P₂)

7

TCN

# of residual blocks (\(\:L\))

2

Dropout rate (\(\:{p}_{c}\))

0.4

Kernel size (\(\:{K}_{T}\))

4

Group dimension (d_group)

16

Dropout rate (\(\:{p}_{t}\))

0.3