Table 1 Review of relevant works.
References | Method | Advantages | Disadvantages | Dataset |
|---|---|---|---|---|
Bayesian approach integrated with CNN networks | More accuracy than RNN and LSTM | Low certainty Bayesian method | ICS | |
Federal learning | More accuracy than GRU, LSTM, and RNN | Unbalanced data set | KDD, NSL-KDD, and CIDDS | |
WCGAN | More accuracy than RF, DT, and SVM | No Dimension reduction | NSL-K, UNSW-NB15 and BoT-IoT | |
Cyber-physical system in SDN | Decrease the occurrence of false alerts by 35–59% | Lack of intelligent feature selection | SDN traffic | |
Blockchain in smart grids | High confidentiality | Blockchain overhead | – | |
Enhanced Firefly Algorithm and CNN | Accuracy was almost 98% | No feature selection | SDN traffic | |
ML-enabled sensor technology integration | Gaussian SVM exhibits higher accuracy | Imbalance and absence of variable selection | Zigbee traffic | |
SMOTE method and decision tree method | Detection of 5 types of attacks | Abstaining from dimensionality Minimization and selecting optimal features | NSL-KDD | |
Five machine learning algorithms | Low latency and error | Not being able to detect all attacks | DER | |
Combination of three decision-trees | More accuracy than SVM, KNN, and DT | Imbalanced dataset and unreduced traffic dimensions | NSL-KDD | |
CNN | More accurate than LSTM | Lack of CNN optimization | UNSW_NB15 and KDDCup 99 | |
PSO algorithms and autoencoders | Appropriate accuracy | Lack of intelligent feature selection | NSL-KDD and UNSW-NB15 | |
Machine learning and signature-based | Low false alarm rate | Memory waste and blacklist time overhead | The dataset includes MITM attacks. |
