Researchers at Institut Polytechnique de Paris, Telecom Paris (INFRES) have created a new computational method that could detect Distributed Denial of Service (DDoS) attacks more efficiently and reliably, TechXplore reports.
DDoS attacks are cyberattacks that overwhelm a server or website with artificial traffic, rendering it inaccessible to normal users.
Advanced Threats Require Advanced Measure
Many users use firewalls, anti-malware software, or traditional intrusion detection systems to protect their websites or server from DDoS attacks.
As per ComputerWorld, the nature of an attack frequently changes during its execution, necessitating a swift and continuous response over several hours or days.
As the primary effect of most attacks is to consume your Internet bandwidth, CW suggests that a well-equipped managed hosting provider possesses both the bandwidth and the necessary appliances to mitigate the effects of an attack.
But it can be challenging to detect these attacks since they are often done with generative adversarial networks (GANs), which are machine learning techniques that can learn to effectively imitate legitimate user requests.
The recent study aimed to develop a novel machine learning-based approach that could increase the efficacy of DDoS detection systems. They proposed an approach based on two distinct models that can be integrated to create a single DDoS detection system.
The first model is meant to figure out if the network traffic coming in is malicious and block it if it is. If not, it is sent to the second model, which decides if it is a DDoS attack. Depending on what this analysis finds, a set of rules and an alert system are implemented.
The Study at a Glimpse
This study investigated how Machine Learning (ML) and Deep Learning (DL) algorithms can be used to detect DDoS cyberattacks. ML and DL algorithms can assist in detecting these attacks. Still, they can also be tricked by attackers employing ML/DL techniques to generate attack traffic that appears to be legitimate traffic.
Long Short-Term Memory (LSTM) neural networks are proposed to detect DDoS attacks with high precision. When tested against adversarial DDoS attacks generated by Generative Adversarial Networks (GAN), the LSTM-based method proved ineffective.
Finally, the researchers demonstrate how to enhance the LSTM-based detection scheme to detect adversarial DDoS attacks with a high detection ratio of 91.75% and 100%.
A Robust Way to Fend Off DDoS Attacks
This research team's proposed DDoS detection tool has significant advantages over previously developed intrusion detection systems.
Particularly, it is reliable and can detect DDoS attacks with high accuracy. It is adaptable and can be customized to meet the specific needs of particular businesses or users.
In addition, internet service providers (ISPs) can also deploy it easily while being protected against standard and adversarial DDoS attacks.
"While previous studies have explored the use of deep learning algorithms to detect DDoS attacks, these approaches may still be vulnerable to attackers who use machine learning and deep learning techniques to create adversarial attack traffic capable of bypassing detection systems," Ali Mustapha, one of the study's authors told TechXplore.
Mustapha and his colleagues' initial tests showed that their system could detect more advanced attacks that were made to trick machine learning algorithms.
The researchers also did a series of tests in real-time to show how valuable their tool could be. They found that the system met the requirements for detecting real-time DDoS attacks by extracting and analyzing network packets quickly without slowing down network traffic too much.
Stay posted here at Tech Times.
