Answers are grounded in this paper and cite their sources.
Customer behavior in e-commerce platforms changes continuously because of promotions, seasonality, platform design updates, supply constraints, economic pressure, and emerging forms of misuse such as refund abuse or promotion exploitation. Static customer segmentation models are often trained once and then reused for long periods, which can cause segment definitions to become misaligned with current transactional behavior. This study proposes an adaptive drift-aware clustering framework that integrates Population Stability Index (PSI) monitoring with incremental K-Means updating to detect distributional changes and refresh customer groups in a controlled manner. A transaction-level retail dataset containing monthly customer aggregates was used to evaluate drift detection and segmentation quality across a twelve-month observation period. The proposed method was compared with static K-Means, quarterly periodic retraining, and incremental K-Means without drift control. Evaluation used Silhouette Score, Davies-Bouldin Index, adjusted cluster stability, drift detection latency, and PSI contribution analysis. The adaptive framework achieved the highest average Silhouette Score (0.487), lowest Davies-Bouldin Index (0.921), and strongest adjacent-window stability (0.786), while detecting major behavioral shifts with lower latency than baseline methods. The findings show that combining statistical drift monitoring with incremental clustering supports more reliable segmentation under evolving customer behavior. The framework is useful for e-commerce analytics teams seeking timely marketing adjustment, customer-risk monitoring, and security-oriented detection of abnormal purchasing patterns
No takes yet. Share an insight, caveat, or question.
Rozi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: