Co-Op Project: Anomaly Detection
Investigators
Team
- Megha Agarwal
Ericsson partners
Research project
Modern telecommunication networks are interconnected and complex, which demands a failure monitoring mechanism involving minimal human intervention. Anomaly detectors are a vital part of building robust distributed software. Working as an early failure detector, they enhance understanding of system behavior, speed up technical support, and improve root cause analysis. The project uses online unsupervised learning hybrid models VAE-LSTM (Variational Auto Encoder- Long Short-term Memory) to detect abnormal behavior over multiple Radio Access Network (RAN) test channels in KPIs such as Downlink/Uplink throughput and CPU/Memory usage.