Modified Variational Autoencoder and Attention Mechanism‐Based Long Short‐Term Memory for Detecting Intrusions in Imbalanced Network Traffic DOI
Oluwadamilare Harazeem Abdulganiyu, Taha Ait Tchakoucht, Yakub Kayode Saheed

et al.

Security and Privacy, Journal Year: 2025, Volume and Issue: 8(3)

Published: May 1, 2025

ABSTRACT The internet and communication industries have grown at a very quick pace, which has caused massive increase in the volume of data network size. This surge given rise to multitude new attacks, posing substantial challenges for security effectively identifying breaches. To counteract these threats, intrusion detection systems (IDS) been created, utilizing technology scrutinize, monitor, analyze traffic ensure conservation availability, confidentiality, integrity. In networks with imbalanced traffic, malicious cyber‐attacks can easily go unnoticed within large volumes regular data. proficiency concealing their presence poses formidable obstacle Network IDS accurately promptly detecting such threats. Despite extensive research efforts, conventional proposed models are faced persistent issues enhancing accuracy lowering false alarm rates, emerging rare zero‐day types. Previous also emphasized problem uneven distribution potentially leading misclassification attacks. As solution problems, this study multi‐model architecture that leverages attention mechanism‐based long short‐term memory (AM‐LSTM) class‐wise focal loss‐based variational autoencoder (CWFL‐VAE), both aimed detect various forms including or while reducing rates computational complexity. CWFL‐VAE was employed handle focusing on minority classes address issue misclassification; AM‐LSTM used classification, Adam gradient descent technique optimize model. system performance assessed using two datasets: NSL‐KDD, benchmark dataset skewed distribution, CSE‐CIC‐IDS2018, featuring is approximately 83% benign cases. CSE‐CIC‐IDS2018 assessing model due its recent release incorporation contemporary attack types, NSL‐KDD functioned as trustworthy benchmark, testing model's implementation against findings literature. showed good low positive rate 0.12%, 99.37% accuracy, 99.23% Similarly, technique's rate, were 94.2%, 0.22%, 92.39%, respectively. According findings, recommended found be competitive terms precision, detection, incidence positives when evaluated existing methods.

Language: Английский

Modified Variational Autoencoder and Attention Mechanism‐Based Long Short‐Term Memory for Detecting Intrusions in Imbalanced Network Traffic DOI
Oluwadamilare Harazeem Abdulganiyu, Taha Ait Tchakoucht, Yakub Kayode Saheed

et al.

Security and Privacy, Journal Year: 2025, Volume and Issue: 8(3)

Published: May 1, 2025

ABSTRACT The internet and communication industries have grown at a very quick pace, which has caused massive increase in the volume of data network size. This surge given rise to multitude new attacks, posing substantial challenges for security effectively identifying breaches. To counteract these threats, intrusion detection systems (IDS) been created, utilizing technology scrutinize, monitor, analyze traffic ensure conservation availability, confidentiality, integrity. In networks with imbalanced traffic, malicious cyber‐attacks can easily go unnoticed within large volumes regular data. proficiency concealing their presence poses formidable obstacle Network IDS accurately promptly detecting such threats. Despite extensive research efforts, conventional proposed models are faced persistent issues enhancing accuracy lowering false alarm rates, emerging rare zero‐day types. Previous also emphasized problem uneven distribution potentially leading misclassification attacks. As solution problems, this study multi‐model architecture that leverages attention mechanism‐based long short‐term memory (AM‐LSTM) class‐wise focal loss‐based variational autoencoder (CWFL‐VAE), both aimed detect various forms including or while reducing rates computational complexity. CWFL‐VAE was employed handle focusing on minority classes address issue misclassification; AM‐LSTM used classification, Adam gradient descent technique optimize model. system performance assessed using two datasets: NSL‐KDD, benchmark dataset skewed distribution, CSE‐CIC‐IDS2018, featuring is approximately 83% benign cases. CSE‐CIC‐IDS2018 assessing model due its recent release incorporation contemporary attack types, NSL‐KDD functioned as trustworthy benchmark, testing model's implementation against findings literature. showed good low positive rate 0.12%, 99.37% accuracy, 99.23% Similarly, technique's rate, were 94.2%, 0.22%, 92.39%, respectively. According findings, recommended found be competitive terms precision, detection, incidence positives when evaluated existing methods.

Language: Английский

Citations

0