Data validation best practices

Responsible AI Data Management

Maria Prokofieva

Lead ML engineer

What we will cover

• Subgroup analysis
• Missing values
• Outlier removal
• Correcting data inconsistencies
• Feature scaling
• Feature encoding
• Dimensionality reduction

Subgroup analysis

Missing data

• Common in large datasets
• Data deletion
• Imputation strategies and model-based approaches
• Subgroup analysis for validation

Outlier removal

• Statistical methods like z-scores and IQR, or robust scaling
• Validate for fair treatment across data segments

Data inconsistencies

• Data quality affects model integrity and reliability
• Data standardization and validation rules
• Subgroup normalization

Feature scaling

• Feature scaling to transform input features
• Validate by checking the distributions among groups

Feature encoding

• Assess encoding effect on outcomes
• Check for biases and information loss
• Check for overfitting
• Employ regularization and dimensionality reduction

Dimensionality reduction

• Reduce input features and preserve essential information
• May create bias
• Use fairness-conscious techniques like t-SNE

Financial advisor

• "Annual income" and "Investment frequency" features
• Adjust outliers and scale
• Subgroup analysis

Let's practice!

Responsible AI Data Management