Refund fraud analytics for an online retail purchases

Online shopping is growing fast across the globe and so are its complexities. Fraud is a complicated phenomenon and its mitigation is critical for running a smooth business. The case considered for the present study is fraud mitigation in return - refund process managed by the customer services of an online retail business. Predictive analytics approach was used to identify early indicators of agent refund fraud - a rare event. The technique used to solve the problem was a Penalised Likelihood based Logistic Regression model. The proposed model allowed the business to select top 5% sample of refund transactions with a higher likelihood of fraud as indicated and queue them for an audit. Implementation of this model resulted in an incremental lift in fraud capture rate.