Customer churn prediction for retail business

Customer churn happens when a customer discontinues his or her interaction with a company. In retail business, a customer is treated to be churned once his/her transactions outdate a particular amount of time. Once a customer becomes a churn, the loss incurred by the company is not just the lost revenue due to the lost customer but also the costs involved in additional marketing in order to attract new customer. Reducing customer churn is a key business goal of every online business. In this project, we have considered data set from UCI Machine Learning repository. This dataset contains records of transactions that occurred between December 1, 2010 and December 1, 2011. This is recorded from an online retail gift store based in United Kingdom. The customers of this company are mainly wholesalers. This data set is pre-processed by removing NAs, validating numerical values, removing erroneous data points. We then perform aggregations on the data to generate invoice based and customer based data sets. A variable churn is attached to each data point. This churn value is determined based on the customer's transactions. Three algorithms are run on this customer aggregated dataset to predict churn value. They are Random forest, Support vector machines and Extreme gradient boosting. A comparative study is done on these three algorithms.