Project Step: Train Baseline Regression and Classification Models

Unit ID: ML-M03-U08 Estimated active time: 45-60 minutes

Why this matters

This is the first modelling evidence in the course.

The goal is not to win a leaderboard. The goal is to create a careful first baseline report.

The idea

You will produce two small comparisons:

1. Regression: median baseline vs linear regression.
2. Classification: most-frequent baseline vs logistic regression.

For each comparison, you should say:

• what target was predicted;
• what features were used;
• what baseline was used;
• what candidate model was used;
• what metric was reported; and
• what the result does not prove.

Predict

Which result would be more convincing?

• A candidate model beats the baseline once, but the learner cannot explain the metric.
• A candidate model beats the baseline and the learner explains the metric and limits.

The second result is better.

Run or inspect

Use the worked notebook first. Then use the practice notebook to repeat the steps with some code left for you to complete.

The important output is not only the metric table. It is your explanation.

Change one thing

If the classification candidate only slightly beats the most-frequent baseline, do not overclaim. Say the result is early evidence and needs stronger validation in later modules.

Practice

Create a short baseline note:

Check and explain

Your note should include this sentence shape:

In this synthetic dataset, the candidate model ______ the baseline, but this does not prove ______.

Takeaway

A first supervised model is evidence to inspect, not a system to trust.