Date of Award
Due to the extensive use and evolution in the cyber world, different network attacks have recently increased significantly. Distributed Denial-of-Service (DDoS) attack has become one of the fatal threats to the Internet, where attackers send massive amounts of packets to the target system to make online systems unavailable to legitimate users. Proper attack detection measurement is crucial to defend against these attacks. This work proposes a deep learning-based model using a contractive autoencoder to detect anomalies. We train our model to learn the normal traffic pattern from the compacted representation of the input data, and then apply a stochastic threshold method to detect the attack. Three renowned IDS datasets have been used for evaluation—CIC-IDS2017, NSL-KDD, and CIC-DDoS2019. We have assessed the results against a basic autoencoder and other deep learning approaches to show our model efficacy. Our results indicate a successful intrusion detection of the proposed method with an accuracy ranging between 93.41% and 97.58% on the CIC-DDoS2019 dataset. Moreover, it achieved an accuracy of 96.08% and 92.45% on NSL-KDD and CIC-IDS2017 datasets, respectively.
Aktar, Sharmin, "Network Intrusion Detection Using a Deep Learning Approach" (2022). University of New Orleans Theses and Dissertations. 3037.