Evolution of Deep Neural Network Architecture Using Particle Swarm Optimization to Improve the Performance in Determining the Friction Angle of Soil

Table 5

PSO-DNN initialization parameters.

Parameters	Explanation	Value and description
N_swarm	Swarm size	50
max_iteration	Number of iterations	30
i	Initial weight	0.5
_damp	Initial weight damping	0.9
C1	Personal learning coefficient	0.4
C2	Global learning coefficient	0.8
VarMin	Number of min neurons	2
VarMax	Number of max neurons	100
N_hiddens	Number of hidden layers	2, 4, 6, 8, 10
Vmax	Velocity max	0.05  (VarMax − VarMin)
Vmin	Velocity min	−Vmax
Fitness function		DNN
Activation function		Tanh
Early stopping	Handling overfitting	True
Cost function		R^2, MAE, RMSE
Data used		Training/validation dataset