SANN - Results - Time Series Tab

Neural Networks

Select the Time Series tab of the SANN - Results dialog box to access the options described here. This tab is only available for time series models. For information on the options that are common to all tabs, see SANN - Results.

In time series problems, the objective is to predict (later) values of a variable or variables from a number of (earlier) values of the same or different variable or variables. In the most common case, a single variable is involved, and a number of sequential values are used to predict the next value in the same sequence (Bishop, 1995).

SANN supports a more general model: the input and output variable(s) do not have to be the same, and the prediction can be more than one step ahead.

Note: To make predictions for time series analyses beyond the data set, i.e., to use predictions of the network as inputs for making more predictions: You can make future predictions only if the variable is projected onto itself, i.e., you must select a target variable but no inputs. Predicting further beyond the data set is a feature that can only be defined for times series analyses in which a variable is projected onto itself. No future predictions can be made beyond the data set unless the variable is projected onto itself (i.e., no inputs selected). In other words, if you have input variables, there is no "meaningful" way of defining a technique by which you can predict beyond the data set.

Projection. Use the options in this group box to create time series predictions for the dependent variable for cases (time steps) within or outside the data set.

Projection length. Specifies how many time steps (cases to predict) should be performed. There is no limit to how long a projection length could be, but it should not go more than a few time steps beyond the maximum number of cases in the data set. Like extrapolation in a regression problem, a projection length that reaches beyond the last case of the data set can produce results that may be unreliable.

Case (starts from). Specify the case from the data set to use as the start of the projection. Note that case numbers that are smaller than Number of steps + steps ahead cannot be predicted by the network since for such cases no complete input information is available.

Example: If your data set contains 50 cases and the Projection length and Case (start from) are set to 20 and 40, respectively, then your predictions start from case 40. Since there are only 50 cases in the data set, the last 10 of the time step predictions will be outside the range of the data set.

Projection graph. Click this button to create a graph of time series projections for the active networks. You can set the length and start point of the predictions using the options Projection length and Case (start from).

Projection spreadsheet. Click this button to create a spreadsheet of time series projections for the active networks. You can set the length and start point of the predictions using the options Projection length and Case (start from).

Note: The output created by the Projection graph is identical to that of Time series graph provided Projection length = number of cases in the data set and Case (starts from) = 1.

Time series graph. Click this button to generate the time series projection graph.

Time series data. Click this button to generate a spreadsheet containing the data from the projection.