Table 2.
Recent works on AI in the IoT security.
| Papers | Year | Objectives/Purpose | Dataset(s) | Algorithm(s) | Accuracy | Task(s) |
|---|---|---|---|---|---|---|
| 7 | 2019 | Authors used two DL models namely FFNN and SNN to classify attacks in IoT network | BoT-IoT | FFNN, SNN | FFNN: 95.1% | C–D |
| 8 | 2018 | Zhou et al., have proposed a model for cyberattack detection in the IoT environment | NSL-KDD | Deep Feed Forward Neural Network + backpropagation | 95% | D |
| UNSW-NB15 | < 95% | |||||
| 9 | 2018 | Authors evaluated the RBM as a ML model to distinguish between abnormal and normal flow traffic | ISCX | RBM | 89% | D |
| 10 | 2017 | Various ML techniques are used, for detecting intrusions based on the classifications applied on the proposed dataset | Synchrophasor dataset | DNN, SVM, RF, DT, NB |
DNN: 79.86% < 80% |
C–D |
| 11 | 2019 | Authors evaluated the performance of several popular ML methods and a DL model in both multi-class and binary classification for detecting intrusions | TON_IoT | Seven supervised Machine Learning methods | CART: 77% | C–D |
| 12 | 2019 | Authors presents an intrusion detection model based on Genetic Algorithm and DBN | NSL-KDD | Genetic Algorithm and DBN | > 99% | D |
| 13 | 2020 | Authors conducted a comparative study of seven DL approaches under two new datasets for intrusion detection | Bot-IoT and CSE-CIC-IDS2018 | RNN, DNN, RBM, CNN, DBN, DBM, DAE | 98.394% | C–D |
| 14 | 2019 | Authors aims at detecting intrusion to the IoT | IoTD | GAN | 20% higher accuracy | D |
| 15 | 2019 | Intrusion Detection System for IoT based on a ML approach | NSL-KDD | DNN | 97% | D |
| 16 | 2019 | Deep learning-based intrusion detection for IoT networks. Authors adopted Bot-IoT dataset, a newly published IoT dataset, they developed a FNN model (feed-forward neural networks) for multi-class and binary classification | Bot-IoT | FNN | 99.41% | C–D |
| 17 | 2020 | In the classification, their proposed method performed well. The proposed technique improves the precision of classification of normal and abnormal of the network traffic | UNSW-NB15 | AC-GAN | 96% | C–D |
| 18 | 2019 | They used four different DL models in IoT networks and discussed their result analysis in terms of accuracy | NSL-KDD | MLP, CNN, DNN, AE | DNN: 99.24% | D |
| UNSW-NB15 | No accuracy value | |||||
| 19 | 2018 | Authors used five ML classification algorithm for the attack classification problem under three datasets | KDD Cup'99 | C4.5, KNN, SVM, NB, RF | C4.5: 99.94% | C–D |
| NSL-KDD cup | KNN: 99.43% | |||||
| GureKDD cup | KNN: 99.08% | |||||
| 21 | 2018 | Authors have adapted an LSTM to detect cyber-attacks. Their experiment was conducted on two datasets | AWID | LSTM | 98.22% | C–D |
| 22 | 2020 | Authors used a DL model for wireless IDS | AWID | FFDNN |
Binary classification: 99.66% Multiclass: 99.77% |
C–D |
| UNSW-NB15 |
Binary classification: 85.48% Multiclass: 74.78% |
|||||
| 23 | 2019 | Authors used LSTM network model, to classify an incoming packet is a part of malicious traffic or normal | Mirai-CCU | LSTM | 99.46% | C–D |
| Mirai-RGU | 100% | |||||
| ISCX2012 | 99.99% | |||||
|
USTC- TFC2016 |
99.99% | |||||
| 24 | 2021 | Three different sources were studied for the performance of the proposed IDS | Bot-IoT | RNN | 99.912% | D |
| CICIDS2017 | 99.811% | |||||
| Power System | 96.822% | |||||
| 25 | 2019 | The authors describe a new IoT dataset that they used for their experiments. Results showed a high accuracy in Binary classification. The prediction output was classified as either normal or attack | BoT-IoT | LSTM | Binary classification: 98% | C |
| RNN | Binary classification: 97% | |||||
| 26 | 2019 | Authors provide a survey for security of IoT Networks | – | Neural Network | 99% | C–D |
| 27 | 2022 | Authors used kernel extreme learning machine (KELM) for classification to detect anomalies in IoT | – | KELM | 99.40% | C–D |
| 28 | 2022 | Authors aims at detecting intrusion in IIoT using classification and detection methods. They used XGBoost (eXtremely Gradient Boosting) model | TON_IoT | XGBoost | 96.5% | C–D |
| 29 | 2021 | Ullah et al. proposed a DL model IDS based on a CNN for multicast and binary classifications | – | CNN | 99.7% | C–D |
| 30 | 2022 | Saba et al. proposed a DL IDS model for anomaly-based IDS based on a CNN technique. The proposed model is capable to detect any aberrant traffic behavior and potential intrusion |
NID (Network Intrusion Detection) Dataset |
CNN | 99.51% | D |
| BoT-IoT | 92.85% | |||||
| 31 | 2022 | For Edge-Based IIoT Security, Guezzaz et al. uses ML techniques to provide a hybrid IDS | NSL-KDD | KNN and PCA (Principal component analysis) | 99.1% | D |
| Bot-IoT | 98.2% | |||||
| 32 | 2022 | Jamal et al., proposed a methodology for malware attacks classification and detection in IoT network. They used ANN | ToN_IoT | DNN | 94.17% | C–D |
| 33 | 2022 | For IoT networks, Basati et al., proposed a novel DNN-based NIDS characterized by a lightweight architecture of NN (neural network) based on PDAE (Parallel Deep Auto-Encoder) | KDDCup99, CICIDS2017, UNSW-NB15 | PDAE | 99.37% | D |
| 34 | 2020 | Ali et al., aimed to study the effects of poisoning attacks on DL based IDSs for heterogeneous wireless communications. The experimental results show that their proposed poisoning attack significantly decrease the DNN classifier performance by about 13%-17% | NSL-KDD | DNN | 84.14% | D |
| 35 | 2023 | Habibi et al., focused on the IoT Botnets attacks detection. They used MLP model which shows high F1-score and accuracy as well as high specificity and sensitivity | Bot-IoT | MLP | 98.93% | D |