Artificial neural networks are computational models inspired by the
biological nerve cells of the human brain. Artificial neural networks resemble the
human brain in two ways, namely knowledge obtained by the network through the
learning process and the strength of the relationship among nerve cells (neurons)
known as synaptic weights to store knowledge. Previous research has tested 12
backpropogation network training algorithms to determine the fastest training
algorithm. The network parameters used are maximum epoch = 10000(104),
learning rate = 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, target error
= 0.001(10-3). It also used the neuron model 10-18-1. Based on ANOVA test with
an alpha level (α) = 5%, the Gradient Descent with Adaptive Learning Rate
(traingda) training algorithm is the fastest training algorithm with an average time
of 0.007485 ± 0.0004782 at a learning rate = 0.9. In this study, testing of 12
training algorithms was carried out to find out the optimal training algorithm in
the 5-8-1 and 5-10-1 neuron models. The learning rate values are 0.01, 0.05, 0.1,
0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0. Based on the ANOVA test with the level of
alpha (α) = 5% in the 5-8-1 neuron model, the optimal algorithm is the Levenberg-
Marquardt (trainlm) training algorithm with an average speed = 0.00664500 at the
learning rate = 1.0. Meanwhile in the 5-10-1 neuron model, the optimal algorithm
is the Levenberg-Marquardt (trainlm) algorithm training with an average speed =
0.00666000 at the learning rate = 0.3.