Regression-based Business Decision Support: Application in Online Retail

Electronic commerce (e-commerce) opens up various growth possibilities, with product returns being one of the major challenges. Product returns prediction can be beneficial in taking business decisions in e-commerce. This research is based on an e-commerce company in India with nearly 25% product returns as a case study. Logistic-regression-based statistical analysis was used to predict the dichotomous dependent variable (product returns). Product size, payment mode, price, quantity, ZIP codes, and states were considered as independent variables. A decision support system was proposed based on the variables and a machine learning approach. A total of 3,187 past orders were used to train the model, which was applied in a one-month pilot study. The model could predict 60% of correct product returns with an overall efficiency of 87%. The case company used these predictions to take preliminary decisions on predicted return orders, saving 17% of the costs associated with product returns.