Transform your AI governance from policy to practice using Viya Workbench built-in SAS procedures designed for trustworthy AI. This practical session demonstrates how to leverage specific SAS procedures to automate governance checks, ensure model fairness, maintain audit trails, and validate AI systems—all without building custom frameworks from scratch. Learn the exact procedures, parameters, and workflows that turn governance requirements into executable code.
You'll learn coding techniques across these tasks:
Fairness & Bias Detection / Mitigation
PROC ASSESSBIAS
- Measures bias across groups (e.g., gender, race)
- Provides metrics like disparate impact
Bias mitigation in modeling PROCs
PROC FOREST, PROC GRADBOOST, PROC SVM
- Use
MITIGATE BIAS statement to reduce bias during training
PROC ASSESS
- Evaluates:
- Accuracy
- Confusion matrix
- Metrics by subgroup
- Helps identify fairness gaps before deployment
Explainability & Transparency
PROC SHAPLEY
- Global and feature-level importance
- Shows how each variable contributes to predictions
- Uses Shapley values to quantify contribution
PROC LIME
- Local (row-level) explanation
- Explains why THIS prediction happened
PROC PARTIALDEPEND
- Shows how model output changes when one variable varies
- Helps visualize feature impact across ranges
(Also uses PROC ASSESS)
- ROC, AUC, lift curves for transparency into performance
Auditability & Governance
PROC REGISTER
- Registers model into SAS Model Manager
- Captures:
- Enables:
- Model cards
- Governance tracking
- Cross-language models (Python, R, SAS)
Monitoring & Drift Detection
PROC CARDINALITY
- Detects structural data drift
- Compares baseline vs. current data distributions
- Flags missing or changing categories
PROC FREQ
- Used to calculate Population Stability Index (PSI)
- Tracks how distributions change over time
Privacy & Security Techniques
PROC TABULARGAN
- Generates synthetic data
- Preserves statistical patterns while protecting PII
MD5 function
- Hashes sensitive values
- Basic data masking technique
