Table 1 A summary of studied IDSs.
Ref. | Challenge | Innovation | Advantage | Disadvantage | Future work |
|---|---|---|---|---|---|
Detecting known/unknown attacks with high accuracy | Random forest and coresets-based clustering algorithms | ✓ Higher accuracy than classical ML ✓ Reducing the processing time while maintaining the quality of big data analysis | ✓ Higher computation time than basic algorithms ✓ The Lack of statistically examining FNR | - | |
Detecting known/unknown attacks on IVCs | Signature/anomaly-based IDSs using ML and the k-means clustering | ✓ Efficiency in the high volume of data ✓ Running on vehicles in real-time ✓ Using optimization algorithms ✓ Low computational-time complexity | ✓ Failure to detect fuzzy attacks effectively ✓ Less efficient in detecting unknown attacks than IDSs designed for one attack | Using unsupervised and online learning methods | |
Detecting types of attacks while solving algorithm complexity and sample dimension problems | SVM-based IDS using optimization algorithms in VANETs | ✓ Classifying a large amount of data ✓ Managing a small sample ✓ Solving optimum local problems ✓ Controlling the complexity of the classification using a penalty function | High mathematical complexity | Using deep learning and real big datasets for training SVM classification | |
Detecting known/unknown attacks such as Botnet, PortScan, DoS, and Brute Force | ANFIS-based KIDS and CNN-based UIDS modules | ✓ Improving the detection rate ✓ Detecting unknown attacks using soft computing techniques | ✓ High detection time for Botnet attacks ✓ Lower detection rate for Brute Force attacks than others | ✓ Using Deep Learning to Improve Security ✓ Optimizing performance using the proposed methods | |
Detecting various attacks by selecting features and handling class imbalance | Random forest and pre-processing methods | ✓ Reduction of resource consumption ✓ Higher performance than others | Limitations in the fast mobility of vehicles | Developing adaptive IDSs to solve the problem of fast mobility | |
Detecting intrusion while saving time and resources and reducing communication overhead | Centralized IDS containing weighted local models using ensemble federated learning and CNN | ✓ Improving accuracy using PSO ✓ Solving the overfitting problem ✓ Managing a large volume of data | Higher FPR than the K-NN algorithm | ✓ Deploying a model with higher performance in Apache Spark and Kafka ✓ Using nature-inspired optimization methods to reduce dimensionality | |
Detecting routing attacks like the black hole | Proposing MVSDS using real-time traffic monitoring and pre-processing methods | Applying without changes to the routing protocols | Lack of statistical analysis of the proposed scheme with other detection models | ✓ Integrating performance metrics with the proposed IDS ✓ Implementing a reaction scheme for routing attacks |
