Comparison of Clustering Algorithms in Market Research Segmentation

This thesis focuses on the use of cluster analysis for segmentation in marketing research. The goal of the thesis is to determine which methods lead to the best results when clustering objects based on typical marketing variables. The input data consists of ordinal scales used in marketing research. These are analyzed using methods suitable for both quantitative variables (hierarchical clustering with Euclidean distance, k-means, k-medians, and Gaussian Mixture Model), as well as methods suitable for nominal variables (hierarchical clustering with IOF similarity measure, Latent Class Analysis, and k-modes). Evaluation criteria used in the thesis show that the best results in analyzing ordinal variables using methods for quantitative data are achieved by the k-means method, while Latent Class Analysis yields the best results for nominal data. In terms of interpretation of the resulting clusters, the k-means method appears to be the most suitable, making it the most efficient segmentation tool in marketing research among the approaches tested.