Improving Customer Segmentation In E-Commerce Using Self-Organizing Maps (Som)

Abstract

Customer segmentation is one of the core activities for any e-commerce business that requires personalization of the approach towards the customer and the improvement of the overall experience for clients. However, classical approaches to customer segmentation, such as K-Means, face certain problems while analyzing high-dimensional and nonlinear data that cannot be effectively addressed using this algorithm. Therefore, the research focuses on analyzing the Self-Organizing Map (SOM) and its applicability to customer segmentation in modern e-commerce organizations. In particular, it will be discussed that SOM allows for working with customer segments in a high-dimensional space without the need for pre-specified clusters. The Brazilian E-Commerce Public Dataset is selected as the basis for conducting the analysis using SOM. A number of algorithms, such as DBSCAN, K-Means, and Gaussian Mixture Models (GMM), are applied for comparison of results. The quality of clusters in terms of clustering accuracy, compactness, and separability of customer segments is assessed through measures, such as Silhouette Score and Davies-Bouldin Index (DBI). Clustering performed through SOM resulted in an almost perfect Silhouette Score of 0.85 and DBI of 0.45. The performance of SOM was better than that of classical methods, such as K-Means, and allowed for the receipt of important information about customer clusters. Therefore, the results support the conclusion that SOM can be successfully used for customer segmentation in e-commerce. Future research may explore hybrid models that incorporate SOM and other methods and real-time customer segmentation in e-commerce.