كنفرانس بين المللی مهندسی صنايع و سيستم ها

صفحه اصلی / دوازدهمین كنفرانس بين المللی مهندسی صنايع و سيستم ها

A Dynamic RFMT-Based Customer Segmentation Framework with Semantic Cluster Alignment

نویسندگان :

Ali Rashedi Gazari¹ Elham Akhondzadeh Noughabi²

1- دانشگاه تربیت مدرس 2- دانشگاه تربیت مدرس

کلمات کلیدی :

Customer segmentation،dynamic segmentation،RFMT،customer behavior،clustering،semantic alignment،CRM،Data Mining

چکیده :

Customer segmentation is useful only when segment meanings remain stable enough to support longitudinal decisions. In non-contractual digital platforms, customer behavior changes monthly, while independently estimated cluster labels are arbitrary and cannot be directly interpreted as behavioral transitions. This study proposes a dynamic RFMT-based segmentation framework that extends recency, frequency, and monetary analysis with tenure and separates monthly pattern discovery from cross-period semantic alignment. Its distinctive contribution is a score-based alignment layer that maps independently generated monthly clusters to a fixed behavioral-state vocabulary, allowing the customer base to evolve while preserving comparable segment meanings over time. After log transformation and robust scaling, monthly MiniBatch K-Means clustering was evaluated on a 36-month panel containing more than three million customers and thirty million customer-month records. A five-state solution-High Value Loyal, Valuable Active, Regular, At Risk, and Fatigued/Low Value-provided an acceptable trade-off between internal clustering quality, temporal comparability, and managerial granularity. Temporal validation showed that High Value Loyal customers had the highest active-state stability probability (0.61), whereas Fatigued/Low Value customers had the highest exit or inactivity rate (0.63). The resulting aligned state sequences provide a scalable and interpretable basis for migration analysis, retention monitoring, and early-warning applications.