Genetic based density peak possibilistic fuzzy c-means algorithms to cluster analysis- a case study on customer segmentation

Abstract

Finding the target consumers for a business depends heavily on its market segmentation approach. Applying clustering analysis to consumer segmentation is one of the most common methods. However, most clustering algorithms can easily fall into the local optimum solution. Besides, it can be challenging to handle noise and outliers and determine the optimal parameters. The initial cluster centers can also affect the clustering result. Thus, this study proposes a clustering algorithm that first employs density peak clustering to obtain the initial cluster centers. Then, the proposed method integrates genetic algorithm (GA) with possibilistic fuzzy c-means (PFCM) algorithm, where GA is used to optimize the cluster centers and the parameters of the PFCM algorithm to overcome the problems above. Using eleven benchmark datasets, the computational results demonstrate that the proposed algorithm can provide better and more robust results in terms of accuracy, adjusted rand index (ARI), and normalized mutual information (NMI) compared to previous clustering algorithms. Additionally, the proposed algorithm is used to segment customers of a retail company in a dataset containing recency, frequency, and monetary (RFM) variables. The clustering result for customer segmentation is also very promising.

Keywords

• Customer segmentation
• Density peak clustering
• Genetic algorithm
• Possibilistic fuzzy c-means algorithm