Intelligent Evaluation and Early Warning of Liquidity Risk of Commercial Banks Based on RNN

Table 4

Training parameters of liquidity risk intelligent early warning model for commercial banks.

Parameter	Parameter value	Parameter definition
Size	[16，16，16，16，16]	Dimensions of network layers
n	5	Network layers
Activation_ function	ReLU	Hidden layer activation function
Learning rate	0.04	Learning rate, the rate of weight update
Loss function	CrossEntropy	Loss function of cross entropy
Scaling_learning rate	0.001	Learning rate change factor (each epoch), the rate of change for each iteration
Weight penalty L2	0	Weight penalty L2 to limit the weight range
Non-sparsity penalty	0	Non-sparsity penalties are designed to minimize the sum of weights for each layer, i.e., sparsity
Sparsity target	0	Sparsity target, the target value of the sum of weights for each layer
Input zero masked fraction	0	Denoising effect of automatic coding to increase anti-noise ability of network
Dropout fraction	0.4	Dropout network improvement
Epoch	3000	Iteration times
Output	Softmax	Output layer activation function