Multiscale Bidirectional Input Convolutional and Deep Neural Network for Human Activity Recognition

Table 5

Parameter exploration process.

Version	Model	Input	Modify parameter	acc_combo
1	MultiConv2D	Origin data	Adam、GlobalMaxPooling2D	0.730
2	MultiConv2D	Origin data	Modify the optimization parameter to rmsprop	0.746
3	MultiConv2D	Origin data	Add the BN layer to the convolution layer	0.786
4	MultiConv2D	Origin data	GlobalAveragePooling2D	0.792
5	MultiConv2D	Forward and reverse sequence	—	0.812