Backpropagation is one of the supervised and popular learning models in
artificial neural networks. Some parameters that affect the performance of the
backpropagation learning algorithm include the number of neurons in the input
layer, neurons in hidden layers (hidden layers), epoch maximum used and the
learning rate. The performance of the training algorithm is said to be optimal can be
seen from the speed produced. The smaller the speed produced, the more optimal the
performance of the algorithm. Tests conducted in previous studies with 10 neurons in
hidden layer (hidden layer) obtained that the most optimal training algorithm based
on the results of the smallest error pelathian Levenberg-Marquardt algorithm
(trainlm) with an average MSE 0.001 with a test level of α = 5%. This research was
conducted to obtain the most optimal algorithm in terms of network speed with
network parameters used in the form of neurons in the input layer, neurons in hidden
layers, neurons in the output layer, learning rate with maximum epoch = 10,000. This
study uses a mixed method, namely development research with qualitative and
quantitative testing (using ANOVA statistical tests). The research data is taken from
random data with 10 input neurons in the input layer, 12 neurons in the hidden layer
(hidden layer) and 1 neuron in the output layer. The results of the analysis in this
study indicate that the Gradient Descent training algorithm with Momentum and
Adaptive Learning Rate (traingdx) is the most optimal algorithm based on network
speed at the learning rate = 1.0 with an average value of 0.007785 ± 0.0004955
seconds.