Academy / Data & Machine Learning

Path 2 · Course 2 · For analysts and aspiring practitioners

Data Quality & Preparation for AI

Profile, clean, label and document data so AI systems learn the right lessons: data quality dimensions, profiling, cleaning decisions, labelling and agreement, representativeness and leakage, and data governance — with hands-on pandas and scikit-learn.

📘 6 modules⏱ About 3.5 hours✅ Knowledge check per module🎓 Final assessment & certificate🎯 Basic Python & pandas

0 of 6 modules complete

Start Module 1 Final assessment

Modules

  1. 1

    Why data quality decides AI quality

    Garbage in, garbage out; six dimensions of data quality; why AI raises the stakes; the preparation workflow; and the practice dataset.

    25 min
  2. 2

    Profiling and exploring data

    Structural profiling, categorical value counts, numerical summaries and outliers, validity checks, visual profiling and data quality reports.

    30 min
  3. 3

    Cleaning data

    Logging decisions, duplicates, standardising categories, fixing types and invalid values, missing value strategies and outliers.

    35 min
  4. 4

    Labelling and annotation

    Why labels matter, labelling guidelines, Cohen's kappa in code, active learning, weak supervision and LLM-assisted labelling, finding label errors, and labellers' welfare.

    30 min
  5. 5

    Representativeness, splits and leakage

    Comparing data with the population served, proxies and historical bias, stratified, time-based and entity splits, data leakage, and leakage-safe preprocessing pipelines.

    35 min
  6. 6

    Documenting and governing data

    Datasheets, lineage and versioning, UK GDPR principles for AI data, pseudonymisation in code, the EU AI Act's Article 10, and continuous data quality checks.

    30 min