Using SAS ® to Build Customer Level Datasets for Predictive Modeling

Content Provider	Semantic Scholar
Author	Shockley, Scott
Copyright Year	2012
Abstract	If you are using operational data to build datasets at the customer level, you’re faced with the challenge of condensing a plethora of raw transactions into a data set that summarizes each customer, one row per customer. You will probably have to use multiple tables with different levels of granularity. Some of the data will change over time, and some of it won't. If the focus of your research is to predict events like customer defection, then changes over time will be a major consideration and make this process even more difficult. The goal of this paper is to guide readers through the process of transforming raw data into a data set for predictive modeling that accurately represents a customer and the factors that could possibly impact the outcome being predicted. The paper will use specific examples like how to calculate derived variables based on complex conditions related to time. For example, it will show how to calculate an average billing amount over time, but only during the most recent uninterrupted period of customer tenure. The discussion mainly concerns the technical details, but also the business and mathematical logic behind the decisions being made.
File Format	PDF HTM / HTML
Alternate Webpage(s)	http://www.scsug.org/wp-content/uploads/2013/11/Using-SAS-to-Build-Customer-Level-Datasets-for-Predictive-Modeling-Scott-Shockley.pdf
Language	English
Access Restriction	Open
Content Type	Text
Resource Type	Article