Abstract:
Forecasting method using Artificial Neural Networks with Backpropagation method starts from
defining the input node, the hidden and output, then data is normalized according to the activation
function is used. Then it is followed by determining the value of learning rate, error tolerance and
maximum iteration values. The network was trained by some of the historical data to approximate
the pattern data. Backpropagation is used to train the network because of the learning process has a
better performance for the method of learning, which is supervised and is usually used by the
perceptron with a lot of layers to change the weights connected to the neurons that exist in the
hidden layer. Each node signal processing with activation function is a rule for determining the
sum of input to output processing elements through choice. Activation function expressing the
relationship between the level internal of activation and output can be either linear or nonlinear.
The purpose of this transformation is to modify the output into the range of values. Activation
function to be used is the Binary Sigmoid function
