Çevrimiçi iadelerin tahmininde müşteri satın alma davranışlarının değerinin entegre bir anomali tespit yaklaşımı ve kupon dağıtımı ile incelenmesi

This study explores the influence of customer behavior on product return rates by applying anomaly detection techniques to customer-level transaction data. Using data from a European e-commerce company, various algorithms are used to identify anomalous transactions, which are then incorporated into a logistic regression model to assess their predictive value for returns, with performance measured by AUC. The best-performing model is used in a second phase to guide coupon distribution strategies, estimating return probabilities under both coupon and non-coupon conditions. These estimates inform heuristic policies within a dynamic programming framework to optimize coupon allocation and evaluate revenue outcomes.