Fig 1.
A typical SDIoT scenario.
Table 1.
Names of each attack and category.
Fig 2.
Process of SMOTE during synthesis and eliminating noisy samples using ENN.
Fig 3.
High-level design of anomaly detection framework in SDIoT.
Fig 4.
Flow diagram of the ML-based anomaly detection model.
Table 2.
Selected features for FSMI model with both datasets.
Table 3.
Performance metrics for the model.
Table 4.
Parameter settings of different models.
Table 5.
Number of Instances (before and after balancing).
Fig 5.
Results of accuracy with increased features.
(a) CIC-IoT-2023 dataset, (b) NSL KDD99 dataset.
Table 6.
Selected features and their description.
Table 7.
Evaluation of SFMI without SMOTE-ENN (in %).
Table 8.
Evaluation of SFMI with SMOTE-ENN (in %).
Table 9.
Evaluation of BFE with SMOTE-ENN (in %).
Table 10.
Evaluation of SMOTE-ENN+SFMI+PCA (in %).
Table 11.
Results comparison with classical machine learning method.
Fig 6.
Multiclass comparison of algorithms on test data for KDDCup 99.
(a) Accuracy, (b) Precision, (c) Recall, (d) F1-Score.
Fig 7.
Multiclass comparison of algorithms on test data for CIC-IoT23.
(a) Accuracy, (b) Precision, (c) Recall, (d) F1-Score.
Fig 8.
Comparison between proposed FS vs classical FS.
(a) KDD Cup99, (b) CIC-IoT 2023.
Table 12.
Testing time (Sec) of different models with feature selection.
Fig 9.
Comparison between SMOTE and SMOTE-ENN in execution time with F-15 and F-10.
Fig 10.
Testing time (sec) of proposed model with PCA vs without PCA.
(a) KDD Cup99, (b) CIC-IoT 2023.
Fig 11.
CPU and Memory usage of both the datasets.
(a) CPU Usage, (b) Memory Usage.