Database Reference
In-Depth Information
Neural Network Methods
Like the regression method, the neural network can be used to forecast
time-series data. As with the regression method, the inputs are lagged
observed values, and the output is a prediction of the next
n
values. For
example, the following test code uses a simple neural network that uses the
previous
k
observed values as its input. The output is a prediction for the
next point in the time series.
The neural network itself is initialized with
k
input units, a hidden layer
containing an arbitrarily chosen five units and a single output layer:
int
k = 14;
double
[] x =
new double
[k];
double
[] y =
new double
[1];
double
t = 0.0;
MersenneTwisterFast twist =
new
MersenneTwisterFast();
NeuralNetwork nn = NeuralNetwork.
build
()
.inputs(k)
.layer(5)
.layer(1)
.initialize();
Inthisexample,simulateddataisgeneratedfromasinewaveandthennoise
is added to the signal:
