Detecting anomalous network patterns

India holds the record for having the highest number of digital transactions annually. VuNet is a major Indian player in this, helping several prominent banks through AI driven monitoring of their payment transaction flows and network infrastructure to improve the user experience. Through our flagship product, vuSmartMaps, we have been rigorously analysing millions of transactions, applications, and network traffic, by collecting, analysing and correlating terabytes of telemetry across their transaction logs, application and system logs and network traffic details to detect and correct failures in real time.

We have extensive experience in analysing various logs and multivariate time series data at scale. Building on this, we have developed a unique approach to anomalies: capturing both transaction anomalies and network anomalies, proactively catching failure incidents, and accelerating root cause analysis through advanced correlation mechanisms. We are also extending the anomaly detection systems to our customer’s network systems to identify spurious network traffic by baselining user and branch network behaviour.

Monitoring more than 2.5 Billion transactions a month across 10K+ network nodes, our anomaly systems have become robust over time to discern various time series patterns from seasonal, multimodal, and sudden spikes. They have been tested against global benchmarks with demonstrated superior results and are constantly enhanced with user feedback loops.

In our talk, we will share our experience around the challenges of varied time series data, a novel way at building anomaly systems and applying to real world noisy data at scale.