Machine Learning methods
■ Semi-supervised:
– Use mechanisms
like AE
– Use labelled
and/or unlabelled
training data
■ Problems:
– Not good enough
performance and
no validation
– Complexity of
algorithms
5
■ Supervised:
– Best accuracy
– Feature selection
– Most reliable
– Measurable
performance
■ Problems:
– Label creation
– Balanced
representation of
all classes
■ Unsupervised:
– Newest methods
– Can use real
traffic for training
(unlabelled data)
■ Problems:
– Not good enough
performance and
no validation
– Complexity of
algorithms