Monitoring Model Drift and Performance in Driverless AI

This lesson walks through monitoring a deployed model in H2O Driverless AI using a credit card fraud example. It covers the metrics tracked after deployment, including prediction volume, average scoring latency, and a drift summary with configurable alert thresholds, illustrated with a distribution chart for a feature that was artificially drifted. It also shows an agentic component that can analyze the training data, generate predictions through a curl request, and compare new predictions against training data using natural language prompts, producing visuals, code, and synthetic data along the way.