Advanced

Return Rate Prediction

Karlén, Erik and Welin, Caspar (2017) FMS820 20171
Mathematical Statistics
Abstract (Swedish)
Product quality is a major concern for all companies. Predicting the lifetime return
rate allows for identification of products with atypically high return rates. Such
products might otherwise not have been detected before it is too late to take any
action to reduce the return rate. Furthermore, predicting a product’s return rate
allows for estimation of the company’s expected cost of handling returns.
This thesis explores multiple methods for predicting a product’s lifetime return rate.
One of the methods investigated utilizes a mixture cure model for the time to return
distribution and includes a parameter for the probability of return. Different time
to return distributions were used in the model; the negative binomial... (More)
Product quality is a major concern for all companies. Predicting the lifetime return
rate allows for identification of products with atypically high return rates. Such
products might otherwise not have been detected before it is too late to take any
action to reduce the return rate. Furthermore, predicting a product’s return rate
allows for estimation of the company’s expected cost of handling returns.
This thesis explores multiple methods for predicting a product’s lifetime return rate.
One of the methods investigated utilizes a mixture cure model for the time to return
distribution and includes a parameter for the probability of return. Different time
to return distributions were used in the model; the negative binomial distribution
and the Weibull distribution. The parameters of the model were estimated using
variations of the Expectation Maximization algorithm.
A simple way of estimating the lifetime return rate is by simply dividing the number of observed returns with the number of observed sales, called the aggregated
return rate. All methods tested outperformed the aggregated return rate and the
most accurate method proved to be the cure mixture model using a negative binomial distribution. (Less)
Please use this url to cite or link to this publication:
author
Karlén, Erik and Welin, Caspar
supervisor
organization
course
FMS820 20171
year
type
H2 - Master's Degree (Two Years)
subject
language
English
id
8919028
date added to LUP
2017-06-27 09:43:04
date last changed
2017-06-27 09:43:04
@misc{8919028,
  abstract     = {Product quality is a major concern for all companies. Predicting the lifetime return
rate allows for identification of products with atypically high return rates. Such
products might otherwise not have been detected before it is too late to take any
action to reduce the return rate. Furthermore, predicting a product’s return rate
allows for estimation of the company’s expected cost of handling returns.
This thesis explores multiple methods for predicting a product’s lifetime return rate.
One of the methods investigated utilizes a mixture cure model for the time to return
distribution and includes a parameter for the probability of return. Different time
to return distributions were used in the model; the negative binomial distribution
and the Weibull distribution. The parameters of the model were estimated using
variations of the Expectation Maximization algorithm.
A simple way of estimating the lifetime return rate is by simply dividing the number of observed returns with the number of observed sales, called the aggregated
return rate. All methods tested outperformed the aggregated return rate and the
most accurate method proved to be the cure mixture model using a negative binomial distribution.},
  author       = {Karlén, Erik and Welin, Caspar},
  language     = {eng},
  note         = {Student Paper},
  title        = {Return Rate Prediction},
  year         = {2017},
}