Skip to main content
Explainability tools in our platform provide in-depth insights into how individual features drive model predictions, offering both global and local perspectives on feature importance and impact. By leveraging SHAP values, users can analyze feature influence across all predictions to identify high-impact features globally, as well as drill down into local explainability for specific instances. These insights help model owners interpret model behavior, diagnose issues such as drift or performance drops, and understand cohort-specific dynamics that may require tailored model adjustments. With explainability, users can maintain transparency, enhance model trustworthiness, and take proactive steps to refine model performance and fairness.

​
Explainability Tutorials

Examples for logging explainability metrics. Click here for more information on how to log feature importance and use explainability.
SHAP: Guide to Getting StartedColab Link
SHAP: Neural Network on Tabular DataColab Link
Surrogate Model ExplainabilityColab Link
One Hot Encoding DecompositionColab Link

​
Explainability Approaches

Arize supports 2 methods for ingesting and visualizing feature importance
f430f18f-image

Method

User Calculated SHAP
171e0488-image

Method

Surrogate Model