Framing a Problem for Machine Learning
Turn a vague business wish into a clear prediction task, or learn that it does not need ML at all
A taste of a lesson
My manager wants AI to predict which customers will leave. Where do I even start?
Start with the action, not the model. What will the team do with the prediction, for example call people or send an offer? Then make 'leave' precise: cancel within the next 60 days, measured from the 1st of each month. Each row is one active customer on that date, and every feature must be known by that date. Finally, set a baseline, such as 'customers with no login for 30 days'. Try this: write your version of the target sentence, including the time window.
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Rewrite a vague goal as a target, a unit of prediction and a prediction moment
- Tell regression, classification, clustering and anomaly detection problems apart
- Set a simple baseline that any model must beat
- Spot when a rule or a process change beats machine learning
- Write a one page problem statement for a data team
Lesson plan
- 1 From wish to decision Identify the decision a prediction would support and who acts on it. Start
- 2 Naming the target and the unit Define exactly what is predicted, for which unit, over what time window. Start
- 3 The prediction moment List only the information available at the moment the prediction is made. Start
- 4 Choosing the problem type Match the frame to regression, classification, clustering or anomaly detection. Start
- 5 Baselines and the no ML option Set a baseline and decide honestly whether a model is worth building. Start
- 6 The one page problem statement Write a complete frame that a data scientist could start from. Start
Try asking
About this tutor
For beginners who have a problem in mind and are not sure whether machine learning fits. You will learn to name the thing you want to predict, decide what information is available at the moment of prediction, tell supervised from unsupervised problems, and set a simple baseline before any model. Lessons use your own example where possible: we rewrite it as a target, features and a decision, then check whether rules, a spreadsheet or a better process would do the job more cheaply. You finish with a one page problem statement you could hand to a data scientist.
Reviews
4.7
3 ratingsSample
- Aisha R.Sample
Free and genuinely useful. It talked me out of using ML for a routing task where three rules did the job.
- Marta G.Sample
I came in wanting to build a churn model and left realising my team first needed a clear definition of churn. The timeline exercise about the prediction moment was the bit that clicked.
- Daniel O.Sample
Very practical. The baseline idea saved me from an embarrassing presentation. I would have liked one more worked example from a non retail setting.
About the teacher
Classical machine learning models, worked through on paper before any code
9 tutors 379 lessons taught Sample
I teach the classical machine learning models: regression, trees, ensembles, nearest neighbours, support vector machines and naive Bayes. My background is in applied analytics, where I spent a lot of time explaining to colleagues why a model made a particular prediction, and that shaped how I teach. I like to start with a tiny dataset you can hold in your...
See Kavya's profile and tutorsMore like this
Other tutors on the same or nearby topics.