Abstract:

In the last decade, analyzing and identifying customers became an irreplaceable need for companies. This research concentrates on discovering a company’s customer segments using different machine learning algorithms, benchmarking different algorithms and its parameters to conclude the best results. Improvements in the technology provided several approaches to dive in and gain insights from a mass amount of data. Machine learning algorithms which is one of the most popular approaches was chosen to convey this empirical study. A dataset with mix categorical and numeric variables is analyzed with one of the conventional machine learning algorithms, namely Hierarchical Agglomerative Clustering Algorithm with Gower’s distance. Kernel Principal Component Analysis is used for preprocessing due to the existence of categorical variables. K-prototypes Algorithm is chosen as benchmark algorithm that fits the qualities of the dataset with mixed categorical and numeric features. Benchmarking provides verification in respect to the accuracy of the results by evaluating the final clusters. Also, examining different parameters and comparing their effects on analysis results indicates the importance and vitality of them for machine learning algorithms, which need to be enlightened to do more accurate analyses. The results showed that both K-prototypes and HAC yield similar results proving that clusters mostly divided appropriately. However, there are a few significant points that are different at both algorithms’ results, which should be examined in further study.

Keywords: