|
Abstract:
|
The problem of data fragmentation is encountered when incorporating certain categorical covariates in a demand forecasting application using survival models. A Bayesian approach is taken to deal with the problem, which enables data fusion across the data segments defined by the covariates. In particular, a nonparametric Bayesian estimator is employed to estimate the survival function of each segment, where the priors are specified by taking advantage of a data segment hierarchy combined with an empirical Bayes method. The resulting hierarchical nonparametric Bayesian models are compared with a class-based modeling approach for demand forecasting using a real data set from a resource-pool-based operation of software development.
|
