A Bayesian Partition Model for Customer Attrition

This paper presents a nonlinear Bayesian model for covariates in a

survival model with a surviving fraction. The work is a direct extension of the

cure rate model of Chen et al. (1999). In their model the covariates depend naturally

on the cure rate through a generalised linear model. We use a more flexible

local model of the covariates utilizing the Bayesian partition model of Holmes et

al. (1999). We apply the model to a large retail banking data set and compare our

results with the generalised linear model used by Chen et al. (1999).