In radius neighbors, outliers are query points that have no neighbors within the specified radius. The behavior when encountering such outliers is controlled
by the outlier handling option, which provides three different strategies:
- none - An error will be thrown when a query point has no neighbors within the radius.
- manual - A specific value will be used for outlier predictions. This value can be customized:
For classification problems, use the
outlier labeloption to specify the label that will be assigned to outlier query points.For regression problems, use the
outlier targetoption to specify the target value that will be assigned to outlier query points.
most frequent - A default value computed from the training data will be used:
For classification problems, the most frequent label among all labels in \(y_{train}\) will be assigned to outlier query points.
For regression problems, the mean of all target values in \(y_{train}\) will be assigned to outlier query points.