Author(s): Chandra Sekhar Sanaboina and A Ramya

Abstract: With the emergence of complex Distributed Denial-of-Service attacks, conventional Intrusion Detection Systems fail to identify novel never-before seen attacks. This project introduces a better model of detection with the assistance of Open-Set Recognition with Reciprocal Points Learning and again enhanced using Attention mechanism. The CNN1D-RPL model is used by the baseline system to accurately detect known and unknown threats based on Euclidean distance mapping in feature space. The extended proposed system extends this further with the addition of an Attention layer to enable the model to dynamically pay attention to salient features, leading to enhanced detection performance. Tested on CICIDS2017 datasets, the extended system performed better than the baseline method, with an accuracy of 99.95%, confirming the fact that it is appropriate for detection of unknown DDoS activity. This combined architecture offers efficient feature extraction, intelligent filtering, and adaptability with incremental learning, and is therefore a valuable tool for real-world cybersecurity applications.

DOI: 10.33545/27075923.2025.v6.i2a.98

Pages: 16-25 | Views: 99 | Downloads: 37

Download Full Article: Click Here