Predictor Space (Nearest Neighbor Analysis)

The predictor space chart is an interactive graph of the predictor space (or a subspace, if there are more than 3 predictors). Each axis represents a predictor in the model, and the location of points in the chart show the values of these predictors for cases in the training and holdout partitions.

Keys. In addition to the predictor values, points in the plot convey other information.

• Shape indicates the partition to which a point belongs, either Training or Holdout.
• The color/shading of a point indicates the value of the target for that case; with distinct color values equal to the categories of a categorical target, and shades indicating the range of values of a continuous target. The indicated value for the training partition is the observed value; for the holdout partition, it is the predicted value. If no target is specified, this key is not shown.
• Heavier outlines indicate a case is focal. Focal records are shown linked to their k nearest neighbors.

Controls and Interactivity. A number of controls in the chart allow you explore the predictor space.

• You can choose which subset of predictors to show in the chart and change which predictors are represented on the dimensions.
• “Focal records” are simply points selected in the Predictor Space chart. If you specified a focal record variable, the points representing the focal records will initially be selected. However, any point can temporarily become a focal record if you select it. The “usual” controls for point selection apply; clicking on a point selects that point and deselects all others; Control-clicking on a point adds it to the set of selected points. Linked views, such as the Peers Chart, will automatically update based upon the cases selected in the Predictor Space.
• You can change the number of nearest neighbors (k) to display for focal records.
• Hovering over a point in the chart displays a tooltip with the value of the case label, or case number if case labels are not defined, and the observed and predicted target values.
• A “Reset” button allows you to return the Predictor Space to its original state.