outlierMeasure

Outlier measure for data in ensemble of decision trees

Syntax

out = outlierMeasure(B,X)
out = outlierMeasure(B,X,'param1',val1,'param2',val2,...)


Description

out = outlierMeasure(B,X) computes outlier measures for predictors X using trees in the ensemble B. The method computes the outlier measure for a given observation by taking an inverse of the average squared proximity between this observation and other observations. outlierMeasure then normalizes these outlier measures by subtracting the median of their distribution, taking the absolute value of this difference, and dividing by the median absolute deviation. A high value of the outlier measure indicates that this observation is an outlier.

You can supply the proximity matrix directly by using the 'Data' parameter.

out = outlierMeasure(B,X,'param1',val1,'param2',val2,...) specifies optional parameter name/value pairs:

'Data'

Flag indicating how to treat the X input argument. If set to 'predictors' (default), the method assumes X is a matrix of predictors and uses it for computation of the proximity matrix. If set to 'proximity', the method treats X as a proximity matrix returned by the proximity method. If you do not supply the proximity matrix, outlierMeasure computes it internally. If you use the proximity method to compute a proximity matrix, supplying it as input to outlierMeasure reduces computing time.

'Labels'

Vector of true class labels. True class labels can be a numeric vector, character matrix, string array, or cell array of character vectors. When you supply this parameter, the method performs the outlier calculation for any observations using only other observations from the same class. This parameter must specify one label for each observation (row) in X.