|
Abstract
|
|
|---|---|
| Multi-dimensional Bayesian network classifiers (MBCs) have been recently introduced to deal with multi-dimensional classification problems where instances are assigned to multiple classes. MBCs have a restricted topology partitioning the set of class and feature variables into three different subgraphs: class subgraph, feature subgraph and bridge subgraph. In this paper, we propose a novel learning algorithm for class-bridge (CB) decomposable MBCs into maximal connected components. Basically, based on a wrapper greedy forward selection approach, the algorithm firstly learns the bridge and feature subgraphs. Then, while the number of components is greater than one and there is an accuracy improvement, it iteratively and sequentially merges together the components, and updates the bridge and feature subgraphs. By learning CB-decomposable MCBs, the computations of MPE are alleviated comparing to general MBCs. Experimental comparison with state-of-the-art algorithms are carried out using synthetic and real-world data sets. The obtained results show the merits of our proposed algorithm. | |
|
International
|
Si |
|
Congress
|
5th European Workshop on Probabilistic Graphical Models (PGM2010) |
|
|
960 |
|
Place
|
Helsinki, Finlandia |
|
Reviewers
|
Si |
|
ISBN/ISSN
|
1458-946X |
|
|
|
|
Start Date
|
13/09/2010 |
|
End Date
|
15/09/2010 |
|
From page
|
25 |
|
To page
|
33 |
|
|
Proceedings of the Fifth European Workshop on Probabilistic Graphical Models |