The net working imitation syaraf can be called intellegance imitation
which like human brain to solve problem. JST is arrengade with algoritma
learning is used to practice net working. In JST is there two learning algoritma.
They are tuition algoritma learning and untuition algoritma learning. One of
tuition algoritma learning is backpropagation. Performance fro this algoritma is
influenced by some net parameter like many neuron in input layer, the number of
neuron in hidden layer, max epoh, the speed higher. The result from neting is
output fasting while do data training. In the research previously have not able
research fiding to 12 algorithm to solve problem. Therefore, this research is done
for 12 algoritm practicing with the statistic ANOVA use alpa (α)=5% to get
output result as feeding every practicing algoritm based on learning rate. This
research uses a mix method. The research data is obtained by conducting
research training data using 12 algorithms 20 times for each learning rate (lr).
The number of neurons in the input layer is 10, the number of neurons in the
hidden layer (n) used is 18 and the output is 1. The value of lr used is 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 are the fastest
training algorithms with learning rate (lr) = 0.9 and the average velocity of
0.007485 ± 0,0004782 produced by the Gradient Descent training algorithm with
Adaptive Learning Rate (traingda) at the level of a (α) = 5%.