Automated Explainability and Bias Testing in DAI
- Observability
- H2O Driverless AI
- H2O Eval Studio
This lesson walks through how Driverless AI automatically generates explainability visualizations once a model is trained, including SHAP values, K-LIME explanations, decision tree surrogates, and both global and individual ICE plots. It shows how SHAP analysis reveals which features drive predictions across a full dataset, while individual reason codes explain a single prediction, such as one a customer service rep might need for a churn decision. The lesson also covers disparate impact analysis, comparing outcomes across privileged and underprivileged groups, and how findings compile automatically into an AutoDoc report.
Key takeaways
- How SHAP analysis and individual reason codes explain both global and single prediction model behavior
- Why individual reason codes matter for a use case like a customer service rep acting on a churn prediction
- What disparate impact analysis measures by comparing outcomes across privileged and underprivileged groups
- How AutoDoc automatically compiles explainability and bias findings into one governance document
Start owning your AI