Comparison of K-Medoids and K-Means Algorithms in Segmenting Customers based on RFM Criteria

Abstract

The company’s approach to customers is important to maintain the company’s profits. Understanding the differences of each customer is very important so that we can understand customer needs based on customer data. Customer Relationship Management (CRM) is considered as a solution to bridge the company and customers. Customer segmentation needs to be done to make it easier for companies to meet customer needs. Data mining and RFM modelling are used for customer segmentation in online retail companies using K-Means and K-Medoids methods. This research compares the performance of both algorithms using Davies Bouldin Index (DBI) and execution time. The results show K-Means is better in cluster validation and execution time. The average DBI value of K-Means is 0.2962 with an execution time of 0.0960 with k=3, while K-Medoids produces a DBI of 0.8942 and an execution time of 2.4295 with k=5. K-Means RFM customer tiers 1-3: Potential Customer/Golden Customer, Lost Customer/ Dormant Customer, and Superstar/Core Customer, 1-5: Champion and Lost. K-Medoids RFM 1-5: Lost, Loyal Customer, Champion, At Risk, and Hibernating, 1-3: Lost Customer/Dormant Customer, Potential Customer/Golden Customer, Superstar/Core Customer, Potential Customer/Golden Customer, and At Risk Customers/Occasional customer.