Towards an Improved Prediction by Preceding a Traditional Classification Method by a Sequence Analysis Method

• Activity Number: 192
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and Marketing
• Author(s): Anita Prinzie*+ and Dirk Van den Poel
• Companies: Ghent University and Ghent University
• Keywords: customer attrition analysis ; classification methods ; decision trees ; Sequence Alignment Method ; sequence analysis ; financial services industry

Abstract

The inability to capture sequential patterns in the data is a typical drawback when using traditional predictive classification methods like logistic regression and decision trees. We overcome this caveat by clustering the observations on a sequential dimension using an element and position-sensitive Sequence Alignment Method (SAM) before estimating a traditional classification model for each of these clusters. Moreover, as we cluster on a sequential dimension assumed to influence the dependent variable, the sequential clustering improves the predictive performance of the overall classification model built on all observations. We illustrate this new procedure, which combines a sequence analysis method with a traditional classification method, by estimating a customer attrition model for a large International Financial Services Provider (IFSP). The results show that, by clustering the customers on their evolution in turnover at the IFSP, which is assumed to influence the customer attrition probability, the predictive accuracy of the overall churn model is improved.