Evaluating Autoencoders Vs Variational Autoencoders For Anomaly Detection In Network Security
Cherla Lavanya Kumari,
Tallapally Mounika,
Akurathi Lakshmi Pathi Rao
Anomaly detection is crucial for maintaining network security, and this study compares the effectiveness
of traditional Autoencoders (AE) and Variational Autoencoders (VAE) for detecting anomalies in
network traffic data. Leveraging their respective architectures, AEs and VAEs are evaluated based on
key performance metrics, including accuracy, precision, recall, F1-score, and AUC-ROC. The results
reveal that VAEs significantly outperform AEs across all metrics, demonstrating higher accuracy (94.0%
vs. 92.5%), precision (92.5% vs. 91.0%), recall (96.0% vs. 94.0%), and F1-score (94.1% vs. 92.5%).
Additionally, VAEs exhibit a superior AUC-ROC of 95.0% compared to 94.2% for AEs. These findings
underscore the VAE's enhanced capability in capturing complex data patterns and distinguishing
between normal and anomalous behaviors more effectively. This study provides valuable insights into
the advantages of probabilistic modeling in improving anomaly detection performance, offering a more
robust solution for network security applications.
