Table 1 Summary for existing model.

¹¹

Employed RF EL in conjunction with correlation FS. For, this system chose 30, 35, and 40 FSs, respectively.

Using two modified classifiers, training them as AdaBoosting and bagging EL, and then combining them using the voting average method.

NSL_KDD has an accuracy of 99.6% with 0.004 FAR, UNSW_NB2015 has an accuracy of 99.1% with 0.008 FAR, and CIC_IDS2017 has an accuracy of 99.4% with 0.0012 FAR.

Due to combination of two EL strategies for separating and spreading legitimate or suspect network traffic attacks, the complexity time measurement took excessively long.

³⁰

N/A

Rule learner-based EL and DT.

The results demonstrate that the IDS classifier techniques have FPR and highest classification accuracy for accuracy, DR, and FAR.

inaccurate findings and multiple attacks that went unnoticed. Additionally, a long period of time spent searching with lowest FNR and accuracy.

³¹

LDA

KNN is used in a two-tier anomaly-detection model.

83.24% accuracy, 4.83% FAR, 82% TPR, 5.43 FPR were evaluated experimentally.

required additional execution time. inadequate handling of anomaly dataset’s network imbalance.

³²

Information gain uses ten features for multiclass and thirty-two features for binary classes.

Adaptive Greedy randomised hybrid RF.

With data gain achieving an accuracy of 78.9035%, the accuracy is 85.0559%.

Less accuracy and high FAR.

³³

DT for FS.

ML methods for hybrid wireless sensor networks that focus on energy efficiency and anomaly detection.

According to the trial findings, the accuracy was 95%, precision was 94.00%, the recall was 98.00%, F1-Score was 96.00%.

A high FAR and a long search time.

³⁴

A genetic algorithm serves as foundation for wrapper method’s feature selection.

For classification, many classifiers are employed.

With 98.75%, 96.64%, and 98.93% DRs, the findings demonstrated accuracy of 98.99% for CIC_IDS17, 98.73% for NSL_KDD, and 97.997% for UNSW_NB15.

inaccurate findings and multiple attacks that went unnoticed.

³³

To choose best subset features, RF was utilized. NSL_KDD, CIDDS-001, and CIC_IDS2017 were utilised.

As a deep learning classifier, the extreme gradient boosting approach is employed.

99% for NSL, 96% for CICIDS-001%, and 92% for CIC_IDS2017 are experimental results.

Because deep learning algorithms are used to separate and disseminate regular or suspect network traffic attacks, measuring complexity time has taken many hours.