Home▸Statistics▸Model Selection: AIC vs BIC

Model Selection: AIC vs BIC

Fit polynomials of increasing degree to noisy data and watch AIC and BIC penalize complexity differently — a live 3D lab for how statistical learning picks a model instead of just chasing the lowest training error.

Statistics3DModerate60 FPS📱 Mobile-adapted⇄ 2D version

Statistical learning isn't just about fitting a curve — it's about choosing which curve to trust. This lab generates noisy samples from a hidden polynomial, fits real least-squares polynomials of degree 1 through 10 to them, and computes the Akaike and Bayesian information criteria for every candidate. A live 3D scene shows the inspected fit overlaying the actual data on one side and a bar chart of AIC (orange) and BIC (green) across all ten degrees on the other, each marked at its own minimum. Push the noise up, shrink the sample, or change the true underlying complexity, and watch how the two criteria disagree on how much model complexity is actually justified by the data.

⚙ Under the hood

Fit least-squares polynomials of degree 1 to 10 to noisy data and watch AIC and BIC penalize model complexity differently, picking a simpler fit than raw training error ever would.

statisticsmodel-selectionregressionaic-bicoverfittingstatistical-learning

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install