Generating
targeted campaigns based on transactional data using Uplift Modeling
Minna Reiman
Abstract
Uplift modeling is a technique to model the
incremental effect of a campaign
or marketing offer. The incremental effect of a campaign is measured by
comparing
two customer groups - one group receiving a
campaign and another
group not receiving any. The goal of this technique is to reduce the
cost of a
campaign by identifying the most optimal targets.
This study presents results from applying uplift models on customer data
at
Klarna, with the goal of understanding how to target
campaigns effectively. To
understand how targeting should be performed, a dataset has been
compiled
based on eight transactional features. The data was then cleaned, preprocessed
and a feature selection was performed based on the Net Information
Value.
Two different models were built - one random forest model based on the
single
model approach, and a gradient tree boosting algorithm based on the
class
variable transformation approach. To evaluate the models, the Qini coefficient
has been used, which is a measure of the discriminatory power of the
model. The conclusion of this study is that for this given dataset, the
gradient
tree boosting algorithm performed more than four times better than the
singe
model, and should therefore be applied to this problem to maximise the
effect
of a campaign.