ROC of multiclass classification in MATLAB

Hi, guys,

I just used the AdaBoost.M2 in a dataset with four-class response variable. I want to produce the ROC curve. The documentation uses the 'plotroc(targets, outputs)' to do it. My question is about the argument of 'outputs'. The documentation says "S-by-Q matrix, where each column contains values in the range [0,1]. The index of the largest element in the column indicates which of S categories that vector presents. ". How to determine the 'outputs' with the results of AdaBoost.M2?

Another question about the '[X,Y] = perfcurve(labels,scores,posclass) '. What is the 'scores' for a AdaBoos.M2 model?

[mehbob ali]

i want to know how you implemented Adaboost.M2

Answer 1

[Alka Nair]

Hi, The PERFCURVE function can be used to plot the ROC for AdaBoostM2. Please see the documentation of function PREDICT, to understand what score referes to for ensemble:

http://in.mathworks.com/help/stats/compactclassificationensemble.predict.html?searchHighlight=adaboostm2%20predict

It is mentioned that, for ensembles, a classification score represents the confidence of a classification into a class. The higher the score, the higher the confidence.

The documentation of PERFCURVE mentions that perfcurve can be used with any classifier or, more broadly, with any method that returns a numeric score for an instance of input data. Please refer to the following page for more information:

http://in.mathworks.com/help/stats/performance-curves.html

[Salma Hassan]

ok if i have 2 columns in score and i determine the class normal , i don't know how to select which columns i should use

[Apoorva Srivastava]

The column that corresponds to the score for the normal class

[Ismat Mohd Sulaiman]

For multiclass, e.g. 3 classes, which one to choose?