Identifying next relevant variables for segmentation by using feature selection approaches

• This paper aims at identifying next relevant variables for segmentation.
• Different feature selection techniques are discussed and applied.
• The performance is assessed using four evaluation metrics for clustering.
• A new approach is proposed and compared with other candidates.
• A real application with data from the concert industry is reported in detail.

Data mining techniques are widely used by researchers and companies in order to solve problems in a myriad of domains. While these techniques are being adopted and used in daily activities, new operational challenges are encountered concerning the steps following this adoption. In this paper, the problem of updating and improving an existing clustering model by adding relevant new variables is studied. A relevant variable is here defined as a feature which is highly correlated with the current structure of the data, since our main goal is to improve the model by adding new information to the current segmentation, but without modifying it significantly. For this purpose, a general framework is proposed, and subsequently applied in a real business context involving an event organizer facing this problem. Based on extensive experiments based on real data, the performance of the proposed approach is compared to existing methods using different evaluation metrics, leading to the conclusion that the proposed technique is performing better for this specific problem.