Neural Network Training
We have time series, i.e., a variable x changing in time xt (t=1,2,...) and we would like to predict the value of x in time t+h.
The prediction of time series using neural network consists of teaching the network the history of the variable over a selected time interval and applying the learned information to the future. Past data are provided to the inputs of the neural network, and future data are expected at the outputs of the network (see figure 2).
As we can see, supervised learning is involved. For more accurate prediction, additional information can be added during training and prediction, for example in the form of interventional variables (intervention indicators) - see figure 3. However, more information does not always mean better prediction; sometimes it can make the process of training and prediction worse. It is always necessary to select truly relevant information, if it is available.
Various types of neural networks can be used for prediction, such as backpropagation, ART, and others. In the rest of this text we will focus on backpropagation.
Figure
2 - Training a time series without interventional variables. The points in the
graph represent a time series obtained by sampling continuous data.
Figure 3 - Training a time series with an intervention indicator
ยท