Skip to content
ML Academy

Classical ML

Interview question

  • midlevel
  • 8 minsuggested time
  • 2Data scientist, Data analyst

You model ancillary revenue per booking with linear regression. Revenue is strongly right-skewed, so you fit it on the log scale and get a good fit. The commercial team then multiplies the predicted values back with exp() and finds that the totals are about a quarter below the actual revenue. What happened and how do you fix it?

Tests: Transformation of the target in regression: interpretation on the log scale, retransformation bias, and alternatives such as smearing or models with a log link.

day 11day 16

Sign in to practise: answers are saved before model answers are shown.