Skip to content
Kudos AI
enfrar

Forecasting and the Split That Lies

Why a shuffled validation split reports a model more than five times better than it is, the naive forecast that beats most methods on a random walk, and what an honest evaluation of a forecaster looks like.

IntermediateModule 330 min · 120 XP
A validation split shuffling until every held-out point sits between training points, the same points cut instead at a single vertical line, and the error bar jumping from 0.76 to 4.29 as the split changes.

This is a premium lesson

Sign in and enrol to read the full lesson, run the code, take the quiz, and earn XP toward the path badge.

Sign inCreate account
← Back to Time Series