Features (Nearest Neighbor Analysis)

The Features tab allows you to request and specify options for feature selection when a target is specified on the Variables tab. By default, all features are considered for feature selection, but you can optionally select a subset of features to force into the model.

Stopping Criterion. At each step, the feature whose addition to the model results in the smallest error (computed as the error rate for a categorical target and sum of squares error for a scale target) is considered for inclusion in the model set. Forward selection continues until the specified condition is met.

• Specified number of features. The algorithm adds a fixed number of features in addition to those forced into the model. Specify a positive integer. Decreasing values of the number to select creates a more parsimonious model, at the risk of missing important features. Increasing values of the number to select will capture all the important features, at the risk of eventually adding features that actually increase the model error.
• Minimum change in absolute error ratio. The algorithm stops when the change in the absolute error ratio indicates that the model cannot be further improved by adding more features. Specify a positive number. Decreasing values of the minimum change will tend to include more features, at the risk of including features that don’t add much value to the model. Increasing the value of the minimum change will tend to exclude more features, at the risk of losing features that are important to the model. The “optimal” value of the minimum change will depend upon your data and application. See the Feature Selection Error Log in the output to help you assess which features are most important. See the topic Feature selection error log (Nearest Neighbor Analysis) for more information.