| Presentation | 2008-03-10 Reconstructing Attribute Dependencies and its Interpretation in Classification Problems : Experimental Results on Modeling Probabilistic Dependencies with Bayesian Network Classifiers Ken Ueno, Youichi Kitahara, Topon K. Paul, Ryohei Orihara, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Recently, many researchers have been focusing on Bayesian Network Classifiers (BNC) as classifiers which can consider dependencies between attributes. In this paper we propose new methods called k-Bayesian Network Classifier 1 (k-BNC1), and k-BNC2 to reconstruct the attribute-dependencies from data, which is based on the conditional mutual information criteria, and the standardized Kullback-Leibler divergence measure, respectively. We compared these classification performances with the performance by the naive Bayes classifier as well as K2 algorithm and Hill Climbing Search algorithm, both of which can automatically construct Bayesian Network structures from data. We also discuss the interpretation of the model generated by our proposed methods, and the advantage of the constraint on directions of arcs between two attributes. Our experimental results show the tendency that k-BNC1 and k-BNC2 perform better than the other methods. The results also suggest that the model with the direction constraints make the classification performance better than the one without the constraints. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Bayesian Network Classifiers / Attribute Dependency |
| Paper # | KBSE2007-59 |
| Date of Issue |
| Conference Information | |
| Committee | KBSE |
|---|---|
| Conference Date | 2008/3/3(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Knowledge-Based Software Engineering (KBSE) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Reconstructing Attribute Dependencies and its Interpretation in Classification Problems : Experimental Results on Modeling Probabilistic Dependencies with Bayesian Network Classifiers |
| Sub Title (in English) | |
| Keyword(1) | Bayesian Network Classifiers |
| Keyword(2) | Attribute Dependency |
| 1st Author's Name | Ken Ueno |
| 1st Author's Affiliation | System Engineering Laboratory, Corporate Research and Development Center, Toshiba Corporation() |
| 2nd Author's Name | Youichi Kitahara |
| 2nd Author's Affiliation | System Engineering Laboratory, Corporate Research and Development Center, Toshiba Corporation |
| 3rd Author's Name | Topon K. Paul |
| 3rd Author's Affiliation | System Engineering Laboratory, Corporate Research and Development Center, Toshiba Corporation |
| 4th Author's Name | Ryohei Orihara |
| 4th Author's Affiliation | System Engineering Laboratory, Corporate Research and Development Center, Toshiba Corporation |
| Date | 2008-03-10 |
| Paper # | KBSE2007-59 |
| Volume (vol) | vol.107 |
| Number (no) | 540 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |