Identifying IoT Devices Based on Spatial and Temporal Features from Network Traffic

Table 6

The structure of Conv-BiLSTM.

Hyperparameters		Value	Activation function
Conv-BiLSTM	Conv2D	#filters = 6, filter size = 5	ReLU
	MaxPlooing2D	#pool size = 2, padding = “valid”	—
	Conv2D	#filters = 16, filter size = 5	ReLU
	MaxPlooing2D	#pool size = 2, padding = “valid”	—
	Flatten	—	—
	Dense	#neurons = 1600, dropout = 0.5	ReLU
	BiLSTM	#neurons = 512, dropout = 0.3	Sigmoid
	BiLSTM	#neurons = 512, dropout = 0.3	Sigmoid
	Dense	#neurons = 18/23	Softmax
Optimizer	—	#Adam with learning rate = 0.001	—
Loss function	—	#Categorical_crossentropy	—
Batch size	—	#512	—
Epochs	—	#50	—