A Dissatisfaction-Driven Approach Using Modeling Customer Purchase Return Behavior in Retail Using SVM

Customer product returns in retail represent a major operational challenge, often stemming from dissatisfaction due to product quality, mismatch with expectations, or service issues. Understanding the underlying behavior associated with purchase returns is critical for enhancing retail strategies and reducing loss. While retailers track return frequencies, they often fail to identify the behavioral patterns and dissatisfaction factors that precede returns. Predictive models rarely include subjective dissatisfaction variables and often underperform on generalization. This study proposes a support vector machine (SVM)-based classification framework to predict purchase return behavior using transactional data, customer feedback, and dissatisfaction indicators such as delivery delay, mismatch in product description, and poor service. The model utilizes labeled return data and applies kernel-based SVM to classify whether a transaction will result in a return. Feature engineering includes behavioral scores, review sentiment scores, product-category mismatch index, and prior return history. Experimental results on a publicly available retail dataset (modified with synthetic dissatisfaction labels) show that the SVM classifier achieved 88.6 % accuracy with an F1-score of 0.85, outperforming baseline models. The analysis highlights that dissatisfaction-related features significantly influence return prediction and can guide proactive interventions by retailers.