Backpropagation is a learning algorithm on Artificial Neural Networks (ANN) and is in great demand for solving problems in various fields of life. Thealgorithm is backpropagation monitored and is usually used by perceptrons with many layers to change the weights connected to neurons in hidden layers. Network parameters affect the performance of algorithms including the number of neurons in the layer input, epoch maximum used and the amount of learning rate. Previous research has tested 12 training algorithms backpropagation. Network parameters in the form of target error = 0.001, maximum epoch = 10000, learning rate = 0.01, with 5 neurons input 10 neurons hidden and one neuron output. The test results produce the algorithm Levenberg Marquardt having the error smallest with level α = 5% and giving an error 0,0001986858. Penelitian previously not discussed at the speed level testing 12 training algorithm. This research was tested on 12 training algorithms 20 times for each learning rate. This research is a mixed method (mixed method), namely development research using quantitative and qualitative methods (using ANOVA statistical tests) at alpha = 5%. The input data uses random data with 10 neurons input, the number of neurons in one hidden layer is 16 and the output consists of 1 neuron with 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 results of this study were that on the neuron model 10-16-1 based on ANAVA test with alpha = 5% it was obtained significantly at 0.786. Based on the descriptive analysis of the Gradient Descent with Adaptive Learning Rate (traingda) is the training algorithm that is the fastest compared to other training algorithms with an average time of 0.007785 ± 0.0005480 seconds and is found in the learning rate = 0.2.