Model Quality & Data Readiness

Learning objectives

• Prepare data for training
• Split data for honest evaluation
• Watch for bias and leakage

Data readiness

A model is only as good as its training data. Clean duplicates, handle missing values deliberately, and keep the training distribution close to the real world the model will see.

Honest evaluation

Never evaluate a model on the data it was trained on. Split into training and held-out test sets so performance numbers reflect real generalization.

Bias and leakage

Leakage occurs when the model sees information during training that it would not have in production - such as the answer to the prediction question. Bias appears when the data under-represents some groups, producing unfair outcomes.

Key takeaways

• Garbage in, garbage out - data quality is model quality.
• Evaluate on data the model has never seen.
• Audit for leakage and bias before deployment.