|
| References | Methods | Advantages | Disadvantages |
|
| Velliangiri et al. [2] | TEHO-DBN | (i) It identified the attack at earlier stages itself | (i) It required more computational time |
| Arul and Punidha [4] | SD-LVQ | (i) It reduced the detection strategies of the DDoS-encrypted cross-site attack | (i) Difficult to process a large amount of data |
| Doriguzzi-Corin et al. [10] | LUCID | (i) Less overhead processing and minimal time of detection | (i) Time of convergence and accuracy were low |
| Agarwal et al. [15] | FS-WOA | (i) It avoided DoS attack entry in a big-scale industry | (i) This method lacks in generating individual instantiations to detect novel attacks |
| Kushwaha and Ranga [3] | SaE-ELM-Ca | (i) It determined appropriate hidden neurons number to improvise model’s learning capability | (i) It failed to utilize multiple connections for testing |
| Alduailij et al. [1] | MI and RFF | (i) It reduced miss classification error | (i) It failed to examine with DL-based detection |
| Alqarni [9] | Ensemble approach | (i) Limited the size of the feature and dataset producing higher performance | (i) Time of execution was high |
| Cil et al. [6] | Feed forward-based DNN | (i) It attained accurate and fast results within a shorter period of time | (i) It preferred the compulsory training process as the large number of packages was contained in the dataset |
| Bovenzi et al. [16] | M2-DAE | (i) It had high efficiency and flexibility | (i) The attack classes were not evaluated |
| Guarino et al. [17] | Machine learning | (i) It obtained high F-measure | (i) More datasets were not considered |
|
