Identification and classification of web traffic inside encrypted network tunnels
The present principles are directed to identifying and classifying web traffic inside encrypted network tunnels. A method includes analyzing network traffic of unencrypted data packets to detect packet traffic, timing, and size patterns. The detected packet, timing, and size traffic patterns are correlated to at least a packet destination and a packet source of the unencrypted data packets to create at least one of a training corpus and a model built from the training corpus. The at least one of the corpus and model is stored in a memory device. Packet traffic, timing, and size patterns of encrypted data packets are observed. The observed packet traffic, timing, and size patterns of the encrypted data packets are compared to at least one of the training corpus and the model to classify the encrypted data packets with respect to at least one of a predicted network host and predicted path information.
1. A system, comprising:
a modeling engine for correlating detected packet traffic patterns, detected packet timing patterns, and detected packet size patterns of unencrypted data packets to at least a packet destination and a packet source of the unencrypted data packets to create at least one of a training corpus and a model built from the training corpus; and
a memory for storing the at least one of the training corpus and the model,
wherein the system further comprises a prediction engine for comparing observed packet traffic patterns, observed packet timing patterns, and observed packet size patterns of encrypted data packets to at least one of the training corpus and the model to classify the encrypted data packets with respect to at least one of a predicted network host and predicted path information for the encrypted data packets,
wherein the at least one of the training corpus and the model is created using at least one random forest.
2. The system of claim 1 , wherein the detected packet traffic patterns, the detected packet timing patterns, and the detected packet size patterns are altered to mimic corresponding features of encrypted data.
3. The system of claim 1 , wherein the at least one of the training corpus and the model is created using a multi-label classification scheme with respect to the at least one random forest, where each label is a prefix of a path for a particular resource or a suffix of a domain name.
4. The system of claim 1 , wherein the at least one random forest comprises a plurality of random forests each having different parameters respectively associated therewith, the method further comprises selecting a best random forest from among the plurality of random forests based on predetermined criteria, and wherein the best random forest is used to provide the at least one of the predicted host name and the predicted path information for the encrypted data packets.
5. The system of claim 4 , wherein the predetermined criteria comprises selecting as the best random forest whichever one of the plurality of random forests has a greatest number of votes for a given label.
6. The system of claim 1 , wherein the at least one of the predicted network host and the predicted path information is determined based on a ranking of possible labels for each of a plurality of input hypertext transfer protocol request and response pairs comprised in the encrypted data packets.
7. The system of claim 1 , wherein the at least one of the predicted network host and the predicted path information is determined based on a mapping of real-valued weights to possible labels for each of a plurality of input hypertext transfer protocol request and response pairs comprised in the encrypted data packets.
8. The system of claim 1 , wherein said modeling engine considers sub-domains and resource paths of the unencrypted data packets when performing the correlating.