Data Science · Advanced level

Advanced level · Module 2 of 6 · 22 guided hours

Predictive Modelling in Data Science Projects

Use predictive models where they help a data science project, evaluate them for the decision at hand and communicate their outputs responsibly.

Learning objectives

  • Decide when prediction rather than description or causation answers the question
  • Build and evaluate predictive models with scikit-learn
  • Use predictions in targeting decisions fairly and transparently

Topics

  • description, prediction and causation
  • scikit-learn workflow recap
  • evaluation for targeting
  • inclusion and exclusion errors
  • fairness in targeting
  • communicating predictions

Lessons

  1. Lesson 1 Description, Prediction or Causation?

    Choose the right kind of analysis for the question and avoid using predictive models to answer causal questions.

    descriptive questionpredictive questioncausal questionproxy variables

  2. Lesson 2 Building a Predictive Model

    Apply the scikit-learn workflow to a targeting problem with proper validation.

    pipelinecross-validationbaseline rulecalibration

  3. Lesson 3 Inclusion and Exclusion Errors

    Evaluate targeting models by who is wrongly included and wrongly excluded, overall and by area.

    inclusion errorexclusion errorcoverage and budgetgroup-wise errors

  4. Lesson 4 Communicating Predictive Results

    Explain predictions, their accuracy and appropriate use to programme staff and communities.

    plain-language performancetransparencyappeals and reviewcommunity feedback

Practical activity

Build a model to predict household food insecurity for targeting, compare it with the current proxy-means rule and analyse inclusion and exclusion errors by area.

Expected outcomes

  • Appropriate use of predictive models
  • Targeting analysis that considers fairness

Assessment

Targeting analysis marked on model evaluation, error analysis and fairness discussion.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.