Random forest based outlier detection
Based on random forest instance proximity measure detects training cases which are different to all other cases.
rfOutliers(model, dataset)
model |
a random forest model returned by |
dataset |
a training set used to generate the |
Strangeness is defined using the random forest model via a proximity matrix (see rfProximity).
If the number is greater than 10, the case can be considered an outlier according to Breiman 2001.
For each instance from a dataset the function returns a numeric score of its strangeness to other cases.
John Adeyanju Alao (as a part of his BSc thesis) and Marko Robnik-Sikonja (thesis supervisor)
Leo Breiman: Random Forests. Machine Learning Journal, 45:5-32, 2001
#first create a random forest tree using CORElearn
dataset <- iris
md <- CoreModel(Species ~ ., dataset, model="rf", rfNoTrees=30,
maxThreads=1)
outliers <- rfOutliers(md, dataset)
plot(abs(outliers))
#for a nicer display try
plot(md, dataset, rfGraphType="outliers")
destroyModels(md) # clean upPlease choose more modern alternatives, such as Google Chrome or Mozilla Firefox.