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

Table 6

Summary of best prediction capability of PSO-DNN models.

Criteria	Number of hidden layers	Best criteria value	Rank	Best neuron structure
R^2	2	0.930	4	(28, 32)
	4	0.926	2	(20, 36, 35, 29)
	6	0.927	5	(17, 44, 52, 52, 38, 28)
	8	0.931	3	(15, 54, 77, 37, 15, 78, 12, 19)
	10	0.938	1	(8, 49, 90, 32, 19, 77, 36, 22, 89, 29)
MAE (^o)	2	1.465	4	(9, 19)
	4	1.426	2	(24, 53, 62, 21)
	6	1.444	3	(9, 76, 27, 13, 56, 77)
	8	1.528	5	(32, 25, 51, 74, 36, 35, 22, 76)
	10	1.395	1	(15, 54, 77, 37, 15, 78, 12, 19, 65, 46)
RMSE (^o)	2	1.925	5	(7, 22)
	4	1.955	4	(23, 36, 34, 41)
	6	1.921	3	(9, 61, 31, 23, 41, 70)
	8	1.839	1	(17, 55, 47, 54, 66, 39, 41, 28)
	10	1.871	2	(15, 54, 77, 37, 15, 78, 12, 19, 65, 46)

Bold values represent the best model for each performance indicator.