Year: 2020
Type: other
Abstract: A data analysis project involving a large data set should be an integral part of any course based on this book.I generally assign a team project to teams of three to four students (the same project to all teams) at just past the midpoint of the course when multiple regression and logistic regression topics have been covered in the class.The project involves applying these two methodologies to develop predictive models for large marketing data sets.Three sample projects are described in the following.They are quite similar in the types of data and the business questions of interest.So the guidelines given below are common to all three projects.Also follow the guidelines for multiple regression model building in Section 6.3.Any special features of the projects are briefly discussed in the individual descriptions of projects.The data sets for the projects are provided at the book's website.Each data set is divided into a training set and a test set (either as two separate data files or as a single file with an indicator variable for the training or test set).Thus, all student teams use the same training set for developing predictive models and the same test set for model validation.So the results of different teams can be directly compared.A 15-20 page report is required from all teams.The response variable in all three projects is the purchase amount by customers in response to a marketing promotion.The predictor variables are their past purchase histories.The response rates are quite small (less than 10%), so a straightforward multiple regression approach does not work since more than 90% of the purchase amounts are zeros, i.e. those customers did not respond to the marketing promotion.In addition, the distributions of the nonzero purchase amounts are highly right-skewed with some unusually large amounts.Therefore, a log transformation of the purchase amount y to y ← ln(y + 1) is recommended.
Author: Ajit C Tamhane
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4231612303