Table2 Summary of Related Work.
From: Robust IoT security using isolation forest and one class SVM algorithms
Author & Year | Detection Model | Targeted environment | Accuracy | Dataset | Noted limitations |
|---|---|---|---|---|---|
Nandanwar et al.9 | 2D-CNN, ResNet | Industry 5.0 | 97.46% | Edge-IIoT-2022 | Not evaluated under adversarial or real-time constraints |
Nandanwar et al.10 | CNN, GRU (AttackNet) | Industrial IoT | 99.75% | N_BaIoT | High compute demands; interpretability not addressed |
Nandanwar et al.11 | Hybrid CNN-BiLSTM + TL | IoT Networks | 99.52% | N_BaIoT | Requires extensive labeled data for transfer learning |
Nandanwar et al.12 | Blockchain-based IDS | IoT healthcare | N/A | N/A | No quantitative performance metrics reported |
Esra et al.13 | DT, RF, kNN, SVM | IoT Networks | ~ 99% + | IoTID20 | Limited to accuracy; no robustness or interpretability study |
Khalid et al.14 | DT, LR, XG Boost | IoT Networks | 94% | UNSW-NB15 | No evaluation under poisoning or adversarial settings |
Imtiaz et al.15 | CNN1D/2D/3D | IoT Networks | 99% + | BoT-IoT, MQTT-IoT-IDS2020, IoT-23 | Not tested on resource-constrained hardware |
Anshika et al.16 | DT, LR, SVM, RF | IoT Networks | RF: 98.47%, SVM: 92.8% | N/A | No benchmark with deep learning or ensemble approaches |
Zeeshan et al.17 | DNN | IoT Networks | 99.01% | IoT-Botnet 2020 | No explainability or poisoning attack analysis |
Dheyaaldin18 | FusionNet | IoMT | 98–99% | WUSTL EHMS, ICU-IoMT | No comparison with simpler unsupervised baselines |
Nadeem et al.19 | RF, ANN, DT, LSTM, AdaBoost, AE | Smart Homes | Up to 100% | UNSW BoT-IoT | Lacks interpretability and adversarial robustness analysis |
Maryam et al.20 | RF, DT, LR, Perceptron, AdaBoost | Healthcare | RF: 99.555% | CIC IoT | No discussion of runtime overhead or edge feasibility |
Lerinaetal.21 | DNN | IoT Networks | 99.89% | N/A | High accuracy, but lacking in ACM real-time or interpretability evaluation |
Abu Al-Haija et al.22 | ELM – survey of variants (S-ELM, U-ELM, Semi-ELM) | Network & IoT intrusion detection (IDS) | Varies | NSL-KDD, CIC-IDS2017, BoT-IoT | Scalability issues on large datasets, potential overfitting, limited handling of multimodal inputs |
Altamimi& Abu Al-Haija23 | ELM | IoT networks (IDS) | NSL-KDD: 99.6% (bin.), 92.5% (multi); Distilled-Kitsune: 99.9% + | NSL-KDD (2009), Distilled-Kitsune (2021) | Limited handling of highly non-linear attacks; requires tuning for real-world use |
