machine-learning evaluation

Definition

Baseline

A baseline is a simple, often heuristic-based model used as a reference point to evaluate the performance of more complex algorithms. A model is only considered useful if it significantly outperforms the baseline.

Common baselines include:

  • Constant Baseline: Always predicts the most frequent class (classification) or the arithmetic mean (regression) of the training set.
  • Random Baseline: Makes predictions based on a uniform or class-proportional distribution.
  • Simple Linear Model: Using linear regression or a perceptron before moving to non-linear models.