Date: 23 May 2025, Friday
Time: 09.00
Place: MA – 205
“Value of Incorporating Customer Purchase Behaviour in Predicting Online Returns:
An Integrated Anomaly Detection Approach and Coupon Distribution”
by
Rana Kaya
(Advisor : Assoc. Prof. Fehmi Tanrısever)
Abstract
This study focuses on the impact of customer behavior on product return rates by using anomaly detection techniques applied to customer-level transactional data. Using a dataset provided by a European e-commerce company, we implement various anomaly detection algorithms to identify transactions that deviate from typical purchasing behavior which are anomaly. The outputs of these algorithms are used to generate anomaly flags, which are subsequently incorporated into a logistic regression model to assess their predictive value for return behavior. Model performance is evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC), and the algorithm which has the highest AUC is selected as the best-performing model. This model, along with a baseline model, forms the foundation for the second phase of the study. In the second phase, we examine the effectiveness of coupon distribution strategies informed by the predicted return probabilities. For each transaction, we calculate the likelihood of return and non-return under both coupon and non-coupon scenarios, using the probability estimates generated by the best and baseline models. These probabilities are then utilized in heuristic policies derived from a dynamic programming framework to optimize coupon allocation decisions. By comparing the revenue outcomes under each model, we aim to evaluate the added value of enhanced anomaly-based predictions in guiding promotional strategies.