|
| Ref. | Model | Data | Eva metrics | Contribution |
|
| [12] | Naive bayes | — | Accuracy | A vast improvement over traditional techniques |
|
| [13] | Nearest neighbour (NN) and linear discriminant analysis(LDA) | WAND | Error rate | Using traffic traces from a variety of network locations, demonstrate the feasibility, and potential of the approach |
|
| [14] | Noise elimination, random forest(RF) | ToN and ISP | Accuracy, F1 | The framework delivers consistently superior performance to other traffic classification schemes in the presence of unclean training data |
|
| [15] | A classifier based on Weka’s classifiers library | — | Recall, precision, accuracy | Authors suggest a fingerprint that is based on zero-length packets, hence enabling a highly efficient sampling strategy |
|
| [16] | NN and SAE | TCP flow data | Precision, recall | The approach solves the problem of nonautomation and poor adaptation in traditional ways |
|
| [17] | CNN | Data including 5 protocol and 5 application | Accuracy | Propose a nearly end-to-end framework for online IP traffic classification |
|
| [19] | LSTM | Real server-generated traffic | Accuracy | The LSTM NNs prove to be a highly efficient computational model capable of solving real server-generated traffic |
|
| [20] | CNN and RNN | Internet of things traffic | Recall, precision, accuracy, F1 | The study shows the performance of CNN and RNN models and a combination of them |
|
| [21] | CNN and LSTM | Mobile traffic | Recall, precision, accuracy, F1 | Introduce two deep learning models for mobile app identification |
|
| [22] | RF, CNN, RNN, multitask learning | QUIC and ISCX | Accuracy | Multitask learning approach out-performs, or performs as accurately as the transfer learning |
|
| [23] | Capsule network | UTSC-2016 | Recall, precision, accuracy, F1 | This study proposed an end-to-end traffic classification method and used the capsule network model for traffic classification |
|
| [24] | CNN, LSTM, SAE | Encrypted mobile traffic | G-mean, accuracy, F1 | This study provided a wide experiment analysis based on multimodel framework (CNN + LSTM) for classification of encrypted mobile traffic |
|
| [26] | RNN, autoencoder | Encrypted traffic | True positive rate, false positive rate, FTF | This study provided the framework containing a multilayer structure which can explore sequential characteristics deeply and import the reconstruction mechanism which can enhance the effectiveness of features |
|
| [25] | CNN, RNN | ISCX VPN-nonVPN | Accuracy, F1 | This study proposed a novel multimodal multitask deep learning approach and DISTILLER classifier, it can solve different traffic classification simultaneously |
|
| [27] | CNN, RNN | MIRAGE-2019 | Accuracy, F1, G-mean, precision | This study used explainable artificial intelligence to improve multimodel behavior, the experiment results showed that the proposed method provide global interpretation, rather than sample-based ones |
|
