Multi-factor evaluation of clustering methods for e-commerce application

This research aimed to investigate the application of Vlse Kriterijumska Optimizacija I Kompromisno Resenje (VIKOR) and Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) multi-criteria decision-making to select the optimal clustering for e-commerce customer segmentation. In this context, clustering as an unsupervised machine learning method offered a way to overcome the limitations of traditional grouping, particularly by providing the ability to capture the diverse needs of consumers. A total of five different clustering methods were considered based on the behavioral data of e-commerce customers. Even though the analyzed algorithms were well-known and widely used, the comprehensive and multidirectional comparison was not trivial. Selected approaches were evaluated on the basis of twelve indicators (decision criteria), divided into four characteristics that take into account both the out-of-context aspects of clustering and the requirements arising from the context of using the clustering results. The results showed consistent outcomes from both analyzed Multi-Criteria Decision Methods, with some notable differences. The methods obtained the same ranking of the top three clustering algorithms (K-median - BIRCH - K-means). However, the TOPSIS and VIKOR sensitivity analysis recommended K-means in 87% of the cases and 60% of the variants verified, respectively. The parameterization of the decision factors had a significant impact on the final ranking of clustering options. This research demonstrated the practical application of the decision methods in selecting the best clustering for multivariate user interfaces to improve personalization in e-commerce.