Ethical AI Compliance Checker

Read Me


Data Preparation Guide for Fairness Audit

To ensure accurate and smooth auditing of your model, please prepare your dataset according to the following requirements:


1. File Format

  • Your data should be in CSV format (comma-separated values).
  • The file must be UTF-8 encoded.

2. Required Columns

Your dataset must include these columns:

Column Name Description Example Values
label The true answer or result your model is trying to predict. 0 (no), 1 (yes)
Protected Attribute The sensitive category you want to check fairness for, like gender, race, or age group. You will provide this column's name when you run the audit. male, female or white, non-white

3. Feature Columns

  • Include all other columns your model uses to make predictions (called features).
  • Features can be numbers or categories.
  • Do not remove or rename the label or protected attribute columns.

4. Data Quality

  • Make sure there are no missing or incorrect values in the label and protected attribute columns.
  • Labels should be simple: only 0 or 1.
  • Protected attribute values should be clear and consistent (e.g., "male" and "female" spelled the same way).

5. Personal Identifiable Information (PII)

  • The system will automatically look for private info like emails, phone numbers, or social security numbers in your data and let you know if it finds any.
  • You don’t need to remove this info before uploading, but be aware it will be flagged for your attention.

6. What the Audit Does

Here’s what happens when you run the audit:

  • It checks your dataset to make sure it has the necessary columns and looks for any private info.
  • It runs your model on the data to see how it predicts outcomes.
  • It compares the model’s predictions to the true answers to measure accuracy.
  • It calculates fairness scores to check if your model treats different groups (like men vs. women) fairly or if it’s biased.
  • It tries to explain which features influence the model’s decisions, helping you understand why it made certain predictions.
  • It saves all these results in a report you can review later.

7. Example Dataset

age income gender label
3455000male1
2862000female0
4558000male1
3960000female0

8. How to Specify the Protected Attribute

When you start the audit, you’ll be asked to enter the name of the protected attribute column (for example, gender, race, or age_group). This must exactly match the column name in your CSV.


9. Troubleshooting

  • If the audit says columns are missing, check your CSV has the label and protected attribute columns.
  • Make sure your CSV file is correctly formatted and readable.

If you follow these simple steps, your dataset will be ready for a thorough fairness check and clear explanations of how your model works.


Upload Model & Data

Past Reports