Baseline

A baseline is the reference point against which a machine learning system is evaluated. It might be a simple rule, a majority-class prediction, an existing production process, a linear model, or the previous model version. A strong baseline reveals whether added complexity actually improves the outcome enough to justify higher cost, latency, maintenance, or risk. The comparison must use the same data split and metrics, and it should include operational measures rather than accuracy alone. Weak or missing baselines can make sophisticated models appear impressive without demonstrating practical value. Baselines should be updated when the environment or current best process changes.