Risk Prediction by Using Artificial Neural Network in Global Software Development

Table 5

Comparison of LM, SCG, and BR of cost-related risks.

	Levenberg–Marquardt	Scaled Conjugate Gradient	Bayesian Regularization
Epoch	6 iterations	49 iterations	238 iterations
Processing	0 : 00 : 00	0 : 00 : 00	0 : 00 : 01
Performance	5.63	5.61	5.05
Gradient	0.00	0.0302	0.0516
MU	0.00		5.00e+10
MSE (validation)	3.09	4.71	1.75
R (validation)	0.94	0.94	0.8524
MSE (testing)	5.78	4.14	6.70
R (testing)	0.92	0.93	0.87