Research Article
Deep Encrypted Traffic Detection: An Anomaly Detection Framework for Encryption Traffic Based on Parallel Automatic Feature Extraction
Table 1
Accuracy comparison of DETD and eight current encryption traffic intrusion detection algorithms.
| Algorithms | Description | Accuracy (%) |
| Yang et al. [17] | Autoencoder | 96.190 | CNN | 97.901 |
| Zeng et al. [18] | 1D-CNN + L1 regularization | 99.851 | LSTM + L1 regularization | 99.222 | SAE + L1 regularization | 98.741 |
| Anderson and Mcgrew [10] | SPLT + BD + TLS + HTTP + DNS | 99.993 | SPLT + BD + TLS + HTTP | 99.983 | SPLT + BD + TLS + DNS | 99.988 | HTTP + DNS | 99.985 |
| Anderson and Mcgrew [11] | Meta + SPLT + BD + TLS + SS | 99.601 | Meta + SPLT + BD + TLS | 99.610 | TLS | 98.202 | Meta + SPLT + BD | 98.900 |
| Anderson et al. [12] | Linear regression | 99.281 | L1-logistic regression | 98.972 | Random forest | 99.990 | MLP | 99.542 |
| Odiathevar et al. [30] | AE + MW + IGMM | 97.435 |
| Kim et al. [31] | RVAE | 97.534 |
| Wang et al. [32] | Model collaboration | 94.353 |
| Our proposed DETD | PASE + RF + adaboost/gradient boosting | 99.994 | PASE + L1_R + adaboosting | 99.998 |
|
|
The bolded values represent the best-performing result.
|