作者: Patrick Haffner , Nicholas Duffield , Jeffrey Erman , Subhabrata Sen , Yu Jin
DOI:
关键词:
摘要: A traffic classifier has a plurality of binary classifiers, each associated with one calibrators. Each calibrator trained to translate an output score the into estimated class probability value using fitted logistic curve, indicating that packet flow on which is based belongs calibrator. The training system configured generate data network information gained and sampling methods. In some embodiments, reduced sets, for class, reducing related not class.