Jun 22

How to calculate lift measure– part II

Lift measures describe the models through the prism of profits which is easily understandable to the management and business users that the model should be applied.

Measure lift determines how the model is better than the absence of models and worse than the best model, but does not tell us how many events (“hits”) we can expect in the first n% of sample sorted by score in descending order. Measures %response and cumulative %response show us just that. Measures as captured cumulative response and response capture tell us how many events exist in the total number of events. All of these measures can be derived by lift measures and give us better understanding quality of the model. Managers like it because they can calculate easily a profit of the model.

The paper describes another lift measures as %response, cumulative %response, captured cumulative %response and %response capture for estimation model using example of model “propensity to buy”. The graphs below are made using SSRS and SQL code is written for MS SQL Server 2012 and higher.

You can see the article here.

Leave a Reply