Machine Learning (ML) · Beginner level

Beginner level · Module 1: Machine Learning Foundations · Lesson 2 of 3

The Machine Learning Workflow

Follow the steps of a real ML project from question to monitored model, and see where most of the effort goes.

In this lesson you will

  • Describe each step of the ML workflow
  • Explain why data preparation and evaluation take most of the time

Key concepts

problem framingdata collection and preparationtraining and evaluationdeployment and monitoring

The lesson

Every ML project follows a similar cycle: frame the problem and the decision it supports, collect and understand data, prepare it, train candidate models, evaluate them on data they have never seen, deploy the best one and monitor it in use.

Most time goes into framing and data. A model trained on the wrong question or on poor data is useless however clever the algorithm. Practitioners often say that better data beats a better algorithm.

The cycle repeats: monitoring may show that the world has changed — new prices, new behaviour — and the model must be retrained. ML is a process, not a one-off product.

Step by step

  1. Read a short case of an ML project.
  2. Map each part of it to a workflow step.
  3. Identify which step the project skipped or rushed.
  4. Discuss what went wrong as a result.
Workflow

The life cycle of a machine learning project

From the decision to support to a monitored model — and back again when the world changes.

  1. Frame the decision

    Which decision will the prediction support, for whom, and what does success look like?

    team + users

  2. Collect and explore data

    Find what exists, check who is represented and what is missing.

    data team

  3. Prepare

    Clean, encode and split the data, keeping the test set untouched.

    data team

  4. Train and compare

    Fit candidate models and compare them with a simple baseline.

    ML practitioner

  5. Evaluate

    Check unseen-data performance overall and per group, in terms of the real decision.

    ML practitioner + users

  6. Deploy and monitor

    Put the model into use with a person reviewing decisions; track performance and drift.

    owner of the model

When monitoring shows the world has changed, go back to data collection and retrain.

Why it works

Framing and data usually take far more time than training — and decide whether the project succeeds.

Practice activity

Draw the workflow for a proposed project — predicting which clinic will run out of a medicine — naming the data and decision at each step.

Expected result: A complete workflow diagram with data sources, the decision supported and how the model would be monitored.

Check your understanding

  1. Why is evaluation done on unseen data?

    Show answer

    To estimate how the model will perform on new cases rather than the ones it learned from.

  2. Why must deployed models be monitored?

    Show answer

    The world changes, so performance can drop over time.

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.