← Lessons

quiz vs the machine

Silver1080

Machine Learning

The Baseline Model First

Ship the simplest honest predictor before reaching for anything fancy.

4 min read · intro · beat Silver to climb

Start dumb on purpose

A baseline is the simplest model that produces a valid prediction. Always majority class, the mean target, last known value, or a tiny logistic regression. It sets the bar every later model must clear.

  • It tells you whether the problem is learnable at all.
  • It exposes pipeline bugs before complexity hides them.
  • It gives a reference point for measuring real gains.

What baselines reveal

If a fancy model barely beats predicting the mean, the features carry little signal. If the baseline is already strong, extra accuracy may not be worth the complexity and serving cost.

  • A human baseline shows the achievable ceiling.
  • A trivial baseline shows the floor.
  • The gap between them is your room to improve.

The flow

Only after a baseline exists does added model complexity earn its keep.

Key idea

A simple baseline reveals whether the problem is learnable, catches pipeline bugs early, and defines the bar that every more complex model must beat.

Check yourself

Answer to earn rating on the learn ladder.

1. What does a baseline model primarily provide?

2. If a complex model barely beats the mean predictor, what is the likely cause?