SAS Trustworthy AI Examples

Overview

This repository contains Python notebooks and SAS programs that demonstrate key principles of Trustworthy AI, including fairness, robustness, and explainability. The examples use publicly available datasets across multiple sectors such as financial services, education, healthcare, and the public sector.

Prerequisites

The examples require access to the SAS® Viya® Workbench.

Installation

Clone the examples repository into the root of your workspace:

git clone https://github.com/sassoftware/sas-trustworthy-ai-examples.git
cd sas-trustworthy-ai-examples

Python Environment Setup

1. Create a Virtual Environment

python -m venv venv

2. Activate the Virtual Environment

On macOS/Linux:

source venv/bin/activate

On Windows:

venv\Scripts\activate

3. Install Required Python Packages

pip install -r requirements.txt

4. Run Examples

Navigate to the python/ directory and open a notebook or script.

Examples

The repository is organized into the following folders:

• data/ — Contains datasets for all 4
• python/ — Contains Python-based Trustworthy AI example
• sas/ — Contains equivalent SAS code examples

Trustworthy AI Examples Covered

Healthcare

• Enhancing Diabetes Risk Predictions with Adaptive Imputation and Monte Carlo Dropout

• Explaining Diabetes Risk Predictions with SHAP and XGBoost

• Improving Heart Disease Risk Predictions with Domain-Adversarial Neural Networks and Ensemble Optimization

• Interpreting Heart Disease Diagnosis with LIME and SVM

• Mitigating Income Bias in Access to Diabetes Screening

• Mitigating Sex Bias in Heart Disease Diagnosis

Finance

• Mitigating Age Bias in Credit Scoring

• Strengthening Loan Amount Estimation with Adversarial Training and Conformal Prediction

• Visualizing Loan Amount Predictions via PDP and ICE

Education

• Adapting Student Performance Predictions with Drift Detection and Adaptive BatchNorm

• Distilling Student Dropout Rules with Surrogate Tree

• Mitigating International Status Bias in Scholarship Allocation

Public/Government

• Mitigating Racial Bias in Employment Service Allocation

• Safeguarding Income Classification with Noise Correction

• Uncovering Income Prediction Logic with RuleFit

Datasets

Example datasets include:

• Adult Dataset
• CDC Diabetes Health Indicators
• German Credit Data
• Heart Disease Dataset
• Student Performance Dataset

View all datasets

Contributing

Maintainers are not currently accepting contributions to this project.

License

This project is licensed under the Apache 2.0 License. See LICENSE for details.

Additional Resources

• Fairlearn - SAS® Trustworthy AI Life Cycle - Trustworthy AI Blog Series - SAS® Viya® Workbench Examples