Artificial neural networks with the backpropagation method are learning methods used for application development. What determines performance on artificial neural networks is the training used. Increased complexity in the number of inputs, maximum epoch used, and maximum speed of understanding. The performance of training algorithms obtained is optimally seen from the errors produced, the smaller the errors produced, the more optimal the performance of the algorithm. In the previous study, the results of errors obtained from 10 input neurons, 14 hidden layer neurons, 1 output neuron with a research level α = 5% were the development of Levenberg-Marquardt training with an average error rate of 0,00010132106600. In this study, 12 training algorithms were tested to study the most optimal algorithms in terms of the smallest error rates using learning rates of 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5 , 0.6, 0.7, 0.8, 0.9, 1. This study uses 2 neuron models, namely 15 input neurons, 26 hidden layer neurons, 1 output neuron and 15 on input neurons, 29 in hidden layer neurons, 1 on the output neuron. The method used is mixed methods (mixed methods), namely development research with quantitative and qualitative testing (using statistical tests). Levenberg Marquardt with the corresponding error value in the neuron model 15-26-1 is 0,00011140701, while in the neuron model 15-29-1 is 0,00012592701. It is expected that at the learning level = 0.8.