Table 13. Comparison of recent intrusion detection approaches on the UNSW-NB15 dataset (training time on whole dataset while testing time on single data sample).
| Study | Method | Performance measures (%) | Multi-class | No of features | Time required (s) | Model size | |||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Feature selection | Classifier | Accuracy | Precision | Recall | F1-score | Training time | Testing time | ||||
| Proposed | RF-RFE | ML ensemble | 98.53 | 98.79 | 98.78 | 98.78 | ✓ | 15 | ~104 | 0.006 | ~544 kb |
| CNN | 98.01 | 97.01 | 98.13 | 98.12 | ✓ | ~413 | 0.08 | ~1,029 kb | |||
| RNN | 98 | 97 | 98 | 97 | ✓ | ~228 | 0.07 | ~250 kb | |||
| LSTM | 98 | 98 | 98 | 98 | ✓ | ~390 | 0.071 | ~978 kb | |||
| Belouch, Hadaj & Idhammad (2018) | CNN | CNN + Dynamic autoencoder | 98.5 | 98.4 | 98.6 | 98.5 | X | – | – | – | ~923 kb |
| Taher, Jisan & Rahman (2019) | IG, PSO, GA |
KNN + RF | 99.96 | 99 | 99 | 99 | X | – | – | – | – |
| Roy et al. (2022) | – | SVM | 93.75 | – | – | – | X | – | ~500 | – | - |
| Dora & Lakshmi (2022) | K Best | RF + XGBoost + DT | 93.7 | 94.5 | 90.2 | 92.29 | X | 20 | ~11.65 | 8 (µs) | – |