Prediction of return in online shopping

Abstract

Mail order business gains popularity day by day. One major problem for the retailers is the high return rates. The incurred cost of returns forces companies to take measures to reduce the number of returns without affecting the customer satisfaction. The aim of this study is to predict whether a purchase results with a return or not based on historical purchase data using machine learning techniques. The data consist of various information about the purchase, customer and the items. Major contribution of this study is to show that using external information to generate features which would otherwise be impossible to extract from data directly improves prediction accuracy significantly.