Novel Hybrid One-Class RNN Model Using Adversarial Data Augmentation for Anomaly Detection in Systems Operations
K. Jaisharma, T Devi., N. Deepa · 2024
The industries are expanding their infrastructure landscape with advanced processors and supercomputers. It is also important to maintain the systems operations robustly and efficiently to handle the service with zero interrupts. But the illegal contractors will be tampering the system's ability by using augmented data dynamically. This research is to detect the anomaly presence through system operations data. The proposed Novel Hybrid One Class RNN model utilizes the advertiser augmented data to compare with the actual system operations data such as processor utilization, data read/write, network utilizations, request/response count, systems temperature and latency measure over the time period. The result shows the Novel Hybrid One Class RNN model achieves the accuracy of 98.4%, precision of 98.1%, recall of 97.8% and F1-score of 97.3% and performed better than the existing models such as K-mean, Gaussian, Markov Chain, Isolation Forest and One Class SVM models for the considered real world AWS EC2 operations dataset collected under various work load conditions.