| Project/Area Number |
13680375
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Statistical science
|
|---|
| Research Institution | Osaka Prefecture University |
|---|
Principal Investigator |
ICHIHACHI Hidetomo Osaka Prefecture University, Graduate School of Engineering, Professor, 大学院・工学研究科, 教授 (30151476)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
HONDA Katsuhiro Osaka Prefecture University, Graduate School of Engineering, Research Associate, 大学院・工学研究科, 助手 (80332964)
|
|---|
| Project Period (FY) |
2001 – 2002
|
| Project Status |
Completed (Fiscal Year 2002)
|
| Budget Amount *help |
¥3,300,000 (Direct Cost: ¥3,300,000)
Fiscal Year 2002: ¥1,100,000 (Direct Cost: ¥1,100,000)
Fiscal Year 2001: ¥2,200,000 (Direct Cost: ¥2,200,000)
|
|---|
| Keywords | Fuzzy clustering / Multivariate data analysis / Information filtering / Data mining |
|---|
| Research Abstract |
In this research, we proposed several hybrid techniques of fuzzy clustering and other multivariate data analysis for finding local features of databases. Our proposed analyzing methods and their properties are as follows.
1. Gaussian mixture model with EM algorithm is a popular density estimation method that uses the likelihood function as a measure of fit. Fuzzy c-Means (FCM) clustering has more flexible structure since the algorithm is based on the objective function method. We showed that just the same algorithm as the Gaussian mixture model can be derived from a modified objective function with regularization by K-L information and the proposed algorithm provided more valid clustering results.
2. We proposed a new approach for the collaborative filtering using local principal components. The new method is regarded as an extension of Fuzzy c-Varieties (FCV) and based on a simultaneous approach to fuzzy clustering with incomplete data sets and Principal Component Analysis (PCA) using lower rank approximation of the data matrix. The missing values are predicted using local linear models derived by FCM-type iterative algorithm.
3. There are two alternatives in fuzzy PCA. One uses fuzzy covariance matrix and FCV is included in this category. The other uses fiizzy correlation matrix. We proposed a linear clustering technique that is the simultaneous approach to fuzzy clustering and PCA of correlation matrix. The simultaneous approach extracts local principal components from incomplete data sets by using the eigenvectors of the correlation matrix.
4. We proposed a modified linear fuzzy clustering method that cannot be easily influenced by outliers using local minor components. The objective function of FCV can be expressed in a simpler form by considering the extraction of local minor components when prototypes are hyperplanes. Using the least absolute deviations, the method provides robust clustering results.
|
