Catch data leakage before training a model
==========================================
Learners choosing a machine-learning foundation

Original practice exercise by AI Course Rank. No purchase required.

STARTING MATERIAL
Invented data: row,customer,month,renewed
1,A,Jan,yes
2,A,Feb,yes
3,B,Jan,no
4,B,Feb,no
5,C,Jan,yes
6,C,Feb,yes
7,D,Jan,no
8,D,Feb,yes
Goal: predict renewal for customers not seen during training.

YOUR TASK
1. Try a row split: train on 1,3,5,7 and test on 2,4,6,8. Identify why that does not test performance on unseen customers.
2. Use customers A and B for training; reserve C and D for testing. Keep the split fixed before comparing any methods.
3. Compute an always-yes baseline on the held-out four rows. Write what this tiny example can and cannot establish.

MY ATTEMPT / OUTPUT
[Write your result here]

TIME SPENT / WHAT WAS DIFFICULT
[Your notes]

ANSWER CHECKS
[ ] The row split shares all four customer identities across train and test.
[ ] The group split shares no customer identities.
[ ] The held-out labels are yes, yes, no, yes.
[ ] The always-yes baseline scores 3/4, or 75%, on this illustrative test.
[ ] Four synthetic examples do not establish real-world predictive performance.

NEXT LEARNING DECISION
If the split and baseline are unfamiliar, choose a foundational course. If they are comfortable and you can code, a practical model-building route may be a better use of time.

Machine Learning Specialization
Compare its structured foundations and prerequisites; check current pricing.
https://aicourserank.com/courses/machine-learning-specialization/
Practical Deep Learning for Coders
Free practical material for learners with coding experience; review the Python and maths prerequisites.
https://aicourserank.com/courses/practical-deep-learning-for-coders/

Some course links may earn a commission. Check current pricing and access terms.
Exercise: https://aicourserank.com/practice/machine-learning-split/