Activity Number:
|
701
|
Type:
|
Contributed
|
Date/Time:
|
Thursday, August 4, 2016 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Statistics in Marketing
|
Abstract #321127
|
View Presentation
|
Title:
|
Using Random Forest to Create Adstock Predictors
|
Author(s):
|
Kathleen Bell* and Scott Dachtyl and Rob Howie
|
Companies:
|
LB Personifi and LB Personifi and Hallmark
|
Keywords:
|
Random Forest ;
Advertising ;
Adstock ;
Variable Reduction ;
variable importance
|
Abstract:
|
A common technique in modeling is to use predictors related to recency and frequency of past behavior. In a non-time series context, the scale of such variables is limited by several constraints - in many cases the p:n ratio can be inappropriately lopsided and large p models can put a significant strain on computing resources. We introduce a novel adstock technique that uses random forest to collapse large series of past behavior into single predictors. The adstock technique preserves a large amount of the information otherwise present in a long string of variables and can function as a powerful single predictor. This methodology can be used to harness the full range of past individual behavior in models.
|
Authors who are presenting talks have a * after their name.