Case Study: Capital One achieves faster incident detection and accurate mobile transaction forecasting with H2O.ai

Capital One Uses H2O for Mobile Transaction Forecasting and Anomaly Detection

Capital One, a Fortune 500 bank with up to 5,000 customers logging into its mobile app per minute, needed a better way to detect volume-based anomalies and forecast user activity—alerts based on simple thresholds missed trends driven by time of day, seasonality, and special events. To solve this, Capital One turned to H2O.ai, leveraging H2O and Sparkling Water to apply machine learning for mobile transaction forecasting and anomaly detection.

Using H2O.ai’s Sparkling Water and H2O algorithms (GBM, grid search) within a cloud-based pipeline on AWS, InfluxDB, and Grafana, Capital One built a scalable production monitoring solution that overlays forecasts with real-time volumes and flags anomalies. The system detected a spike of ~20,000 unexpected logins and alerted ops two minutes after the spike and four minutes before other alarms, identified ~12% missing events after a platform change, and overall improved incident detection times and forecasting accuracy.

Capital One

Rahul Gupta

Data Engineer