همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Customer Segmentation in Online Tire Sales Using RFM and Quantity: A Comparison of K-Means, MiniBatchKMeans, and DBSCAN with Quality Improvement via Density-Based Refinement

نویسندگان :

Seyed Mohammadreza Jalalian Shahri¹ Seyed Hashem Mohtashami²

1- کارخانه کویرتایر 2- کارخانه کویرتایر

کلمات کلیدی :

Online sales،customer clustering،RFM،K-Means،, MiniBatchK-Means،DBSCAN،HDBSCAN،PCA،cluster evaluation

چکیده :

Abstract—Customer segmentation remains a critical challenge in e-commerce, where diverse purchasing behaviors demand robust analytical frameworks. This study introduces a practical approach based on five years of real online sales data from a tire manufacturing company. Behavioral features were derived from the RFM–Qty model (recency, frequency, monetary value, average order amount, and purchase quantity). Three clustering algorithms—K-Means, MiniBatchK-Means, and DBSCAN—were systematically compared. Internal evaluation metrics indicated k = 8 as the optimal cluster number. DBSCAN was subsequently applied as a refinement stage to filter out approximately 3.28% of noisy data points, after which reapplying K-Means on the cleaned dataset led to significant improvements (the silhouette coefficient increased from 0.3876 to 0.3967, and the Davies–Bouldin index decreased from 0.9182 to 0.9022). The final solution yielded eight actionable customer segments with clear behavioral profiles and managerial implications, such as loyalty programs for “Champions,” reactivation campaigns for “At-risk” customers, and strategies to convert “One-time buyers” into repeat purchasers. A two-dimensional PCA visualization capturing about 82% of the variance was employed to illustrate cluster separation. The findings highlight that incorporating density-based noise filtering prior to clustering enhances segmentation effectiveness and supports more targeted marketing decisions in real-world e-commerce environments. This work contributes to both academic research on customer analytics and the practical design of data-driven marketing strategies.