|
Activity Number:
|
16
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, August 6, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #306233 |
|
Title:
|
Modeling Multivariate Time Series with Application to Software Defects Data
|
|
Author(s):
|
Mihaela Serban*+ and Wanli Min
|
|
Companies:
|
Carnegie Mellon University and IBM T. J. Watson Research Center
|
|
Address:
|
5000 Forbes Ave., Pittsburgh, PA, 15213,
|
|
Keywords:
|
EM ; HMM ; Poisson ; multinomial
|
|
Abstract:
|
We develop a Poisson hidden Markov model for categorized data with application to software defects data. The numbers of defects are modeled as a state-dependent Poisson process and, conditioning on the total number of defects we specify, a state-dependent multinomial distribution to characterize the defects by type. We fit two-regimes and three-regimes HMMs using an EM algorithm. Model selection is carried out using BIC for in-sample comparison and mean relative error for out-of-sample comparison. We also compare our model with the NHPP model for categorized software data introduced by Liu, Z. et al. (2005).
|
