Dual-stage autoencoder and density clustering for unsupervised cloud intrusion detection
Abstract
Cloud intrusion detection remains challenging because modern network traffic is high-dimensional, highly imbalanced, and continuously evolving. Although autoencoders and density-based clustering have been applied independently for intrusion detection, directly clustering the learned latent space often reduces the separation between benign and malicious traffic, particularly for minority attacks. This paper presents a dual-stage unsupervised intrusion detection approach that combines latent representation learning with anomaly-filtered density clustering. A dual-function autoencoder is first trained using benign traffic to learn compact latent representations while identifying anomalous flows through reconstruction error. Instead of clustering the complete latent space, only the latent representations of detected anomalies are forwarded to density-based spatial clustering of applications with noise (DBSCAN) for attack grouping. This anomaly-filtered clustering strategy reduces the influence of dominant benign traffic and improves the discrimination of attack patterns in the latent space. The proposed method was evaluated on the CSE-CIC-IDS2018 benchmark dataset containing diverse cloud network attacks. Experimental results achieved 99.46% anomaly detection accuracy and 98.81% multiclass attack classification accuracy using a 21-dimensional latent representation. Improved clustering quality was further confirmed by a Silhouette score of 0.9859 and a Davies-Bouldin index (DBI) of 0.0093. The proposed pipeline provides an effective unsupervised solution for cloud intrusion detection without requiring attack labels during model training.
Keywords
Anomaly detection; Autoencoder; Cloud intrusion detection; Density-based spatial clustering of applications with noise; Latent representation; Unsupervised learning
Full Text:
PDFDOI: https://doi.org/10.11591/eei.v15i6.14121
Refbacks
- There are currently no refbacks.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.
Bulletin of Electrical Engineering and Informatics (BEEI)
ISSN: 2089-3191
,
e-ISSN: 2302-9285
This journal is published by the
Institute of Advanced Engineering and Science (IAES)
.