Network-based technologies have become increasingly popular, so much so that they are now being used by numerous individuals, professionals, and businesses globally. Although these technologies have several advantages, on the downside, most network-based systems are also highly susceptible to malicious attacks. The consequences of such an attack on network-based systems can cause great financial harm and can be extremely severe and devastating. For instance, an attack on the power utility network of a company could leave millions of individuals and offices bereft of electricity. In addition, attacks on social media networks such as Facebook and Instagram can lead to data breaches.
To tackle these vulnerabilities of network-based systems, computer scientists all around the globe have been investing their time in the development of advanced IDSs (Intrusion Detection Systems). These systems could be the key for identifying and fighting against malicious attacks, thus, boosting a network’s safety. In current years, machine learning (ML) algorithms have shown great potential for automatically detecting intrusions and attacks on a network's functioning.
To further advance this concept, researchers have now developed a new feature selection method that would facilitate the development of more efficacious ML-based IDSs. The innovative method would help boost the Intrusion Detection and Prevention Systems Market as the feature proved to work remarkably well in contrast to other commonly employed feature selection techniques available at present.
The most critical part of developing and training ML-based IDSs is a selection of such data features that a model could depend on or focus on while making predictions. Theoretically, researchers could be able to get the most suitable feature for solving a given task via ML tools by analyzing large datasets. Similar is also applicable for intrusion detection.
However, since that is not the case, the present study looked into feature selection within network traffic data with the focus of detecting potential attacks. The research considered several existing feature selections while also proposing a new feature selection algorithm to identify the most promising feature when it comes to network traffic data.
The team proposed a new and distinctive feature selection that can address the challenge of constant input features and discrete target values. Further, they showed that their developed method performs rather well against the standard selection methods.
Researchers identified this feature salient for intrusion detection and were able to build a highly efficient ML-based detection system through this feature. The system demonstrated its capability of distinguishing between DDoS (Distributed Denial of Service) attacks and simple harmless signals with 99 percent accuracy. In the future, the feature selection method that was developed could lead to the creation of highly effective IDSs. Furthermore, the systems created could be implemented in real-world settings to detect malicious attacks on actual networks.