The Use of Artificial Neural Network for Prediction of Dissolution Kinetics

Table 3

Comparison of various backpropagation algorithms using different transfer function (test).

Training function	Transfer function	One hidden layer			Two hidden layers
		𝑅^2	RMSE	MAE	𝑅^2	RMSE	MAE
LM backpropagation	tansig	0.9933	0.0121	0.0099	0.819	0.0641	0.0374
LM backpropagation	logsig	0.9921	0.0135	0.0110	0.9975	0.0073	0.0061
LM backpropagation	radbas	0.4391	0.1320	0.0990	0.5056	0.4184	0.3963
Bayesian regulation backpropagation	tansig	0.9945	0.0113	0.0091	0.9938	0.0120	0.0098
Bayesian regulation backpropagation	logsig	0.9854	0.0167	0.0133	0.9951	0.0102	0.0081
Bayesian regulation backpropagation	radbas	0.9946	0.0117	0.0101	0.9965	0.0090	0.0069
BFGS quasi-Newton backpropagation	tansig	0.9876	0.0167	0.0116	0.9909	0.0136	0.0120
BFGS quasi-Newton backpropagation	logsig	0.9839	0.0189	0.0116	0.9882	0.0150	0.0128
BFGS quasi-Newton backpropagation	radbas	0.0724	0.2041	0.0981	0.5832	0.1397	0.0688