Researchers from Saudi Arabia developed a neural network to detect botnet intrusions in internet networks. The created algorithm showed great efficacy, recognizing malicious traffic with up to 99.77% precision.
As Gazeta.SPb reports, the number of devices connected to the internet keeps rapidly growing. Smart cameras, speakers, home appliances, sensors, and medical systems form a massive environment for potential cyber assaults. Attackers frequently group such devices into botnets—networks of infected hardware used for data theft, server assaults, and disruptions to crucial infrastructure operation. Regular cybersecurity methods frequently prove insufficiently capable. Therefore, specialists from Saudi Arabia chose to discover a remedy for this issue.
As part of the work, specialists utilized the open BoT-IoT dataset, which contains millions of records of real internet traffic with various attack scenarios. According to Scientific Reports, approximately 99.5% of the data in the set pertained to harmful traffic, while the portion of normal activity was under 0.5%.
The system devised by the academics combines two types of neural networks at once. A Convolutional Neural Network (CNN) is responsible for spotting spatial patterns in the data, and an LSTM network analyzes changes in traffic characteristics over time. Because of this, the algorithm is able to efficiently pinpoint even complex distributed assaults, which often go unnoticed by other analysis techniques.
About the Author
