Mixture of Statistics & AI: Comparative Analysis of 27 Algorithms for Marketing

This study attempts to identify the most appropriate clustering algorithm for marketing and other fields that use them because identifying and classifying customers based on their needs and behaviors is essential, which is interpreted by big data science. However, before this, we must implement appropriate segmentation among customers. In this study, based on the opinions of famous faces such as Kotler, Porter, and Drucker about segmentation, information from 460 samples was collected in the form of seven variables by surveying consumers using standard scientific questionnaires. With the use of routine statistics, the validity and reliability of the data were examined and confirmed. Then, with two data mining platforms, Weka and RapidMiner, data clustering was implemented with thirty algorithms at different ranges. However, to compare with other studies, statistics were used more for analysis, and due to the non-normality of the data, non-parametric statistical methods such as Kruskal-Wallis, which is the non-parametric equivalent of the ANOVA test, and DSCF as a post-hoc test were used in addition to a Silhouette index. Finally, based on the statistical index of effect size, which is used in meta-analysis studies to compare results, the outputs were examined and proved that, except for the farthest-first algorithm, others do not show significant values in the seven mentioned variables. This means that for all data that have an abnormal shape on the Likert scale, the arrangement from the outside to the center of the cluster seems to be the most appropriate method for classifying customers.