How good is your model?

auditor: an R Package for Model-Agnostic Visual Validation and Diagnostics

Alicja Gosiewska & Agnieszka Ciepielewska

11 XII 2018, Wrocław

Work on this package was financially supported by the

‘NCN Opus grant 2016/21/B/ST6/02176’.

Why do we need a model audit?

Use case dragons data

The story of 3 models

linear model

support vector machines

random forest

Variables of Linear Model

Linear Model

Linear Model

Linear Model

middle ages

ancient

early modern

Adding period

Adding period

A few years later

RMSE of models

913

5650

1720

12800

2420

6250

lm

svm

rf

height

height

2

Only linear model used the new variable properly

alicjagosiewska@gmail.com

agnieszkac27@gmail.com