Backpropagation is a learning algorithm that exists in Artificial Neural
Networks (ANN) and is much in demand to solve problems. The backpropagation
algorithm is monitored and is usually used by perceptrons with many layers to
change the weights connected to the neurons in the hidden layer. In the
backpropagation method there are 12 training algorithms. Therefore, in this study
20 training algorithms were tested for 20 repetitions of each learning rate (lr) to
get the fastest training algorithm. This study uses a mixed method, namely
qualitative and quantitative methods (using ANOVA statistical test) at the level of
alpha (α) = 5%. Input data uses random data with 15 neurons input 22 neurons in
hidden layer and 1 output and 15 neurons 25 neurons in hidden layer 1 output using
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. The conclusion
of the research is ANOVA statistical test on neuron models 15-22-1, the most
optimal algorithm is obtained in terms of network speed, namely Resilent
Backpropagation (trainrp) training algorithm with an average speed of 0.0068650
at the learning rate = 0.8. Meanwhile on the neuron model 15-25-1, the most
optimal algorithm was obtained in terms of network speed, namely the training
algorithm Levenberg-Marquardt (trainlm) with an average speed of 0.0070500 at
the learning rate = 1.