Skip to content
Predict.ai

Accuracy & evaluation

Baseline forecast

A baseline is a simple rule you must beat: last value, last week, last year, a seasonal naive, a moving average. If a fancy model cannot beat it on a fair backtest, keep the rule and save the drama.

Baselines are not dummy work. They are the price of admission. Vendors who will not publish vs-baseline numbers are selling a chart.

Pick the naive that matches the clock. Last-year-this-week is a high bar in retail. Last-hour is a high bar in energy. Using the wrong naive makes everyone look like a genius.

A baseline should run in production too, as a watchdog. When the champion loses to last week for three origins in a row, something is stale.

One series, four layers

Time series decomposed into trend, season, and leftoverObservedTrendSeasonLeftover
Most business series are a slow drift, a repeating clock, and leftover noise. Models differ in how explicitly they split these. You still want to see the split.

Read next

Related terms

Try it on your data.

Connect an outcome and its history. Predict.ai finds the drivers, races the models, and keeps the forecast live — in the workspace, over the API, and through agents.