A Comparative Analysis of Convolutional, Sequential and Their Hybrid Models in Detecting Cyber-Attacks
Lakshmi S., Smitha Dharan · 2024
There has been a considerable leap in the number and sheer variety of cyber-attacks which has raised the concerns of businesses and other organizations. There has been a constant need for continuously improving the quality of cyber-attack detection systems because of this increasing diversity and complexity in cyber-attacks. The most challenging part in any such system is to detect real time cyber-attacks and this is where most systems fail. Therefore, as part of solving this problem, it would be necessary to design an intelligent system which will detect real time cyber-attacks efficiently. Keeping this in mind, several methods that employ machine learning have been applied with generally positive outcomes. Deep learning, a subfield of machine learning, is being used in attack detection these days with particularly good results. This paper investigates the possibility of designing and implementing a system which uses a combination of convolutional neural networks and a sequential model in detecting cyber-attacks. This work also performs a comparative analysis of the performances of this hybrid model with the convolutional and the sequential models used to design the hybrid model. A comparison of these three models with some traditional machine learning algorithms is also done. The sequential model explored here is Gated Recurrent Unit, a variant of recurrent neural network, which is generally used in sequential data processing. The results obtained after experiments demonstrate that the hybrid method gives considerably strong results in cyber-attack detection and also that this can be a new research direction in this area. It is also proved that the results obtained by the hybrid model are comparable to those of convolutional model and that the sequential model slightly outperforms the other two models.