2002 Fiscal Year Final Research Report Summary

Study on Data Mining by Hybrid Modeling

Research Project

Project/Area Number	13680536
Research Category	Grant-in-Aid for Scientific Research (C)
Allocation Type	Single-year Grants
Section	一般
Research Field	社会システム工学
Research Institution	Meiji University
Principal Investigator	OOTAKI Atsushi Meiji University, Mechanical Engineering Informatics, School of Science and Technology, Professor, 理工学部, 教授 (20061971)
Project Period (FY)	2001 – 2002
Keywords	CART / MARS / AR model / missing data / residual analysis
Research Abstract	To mine data set for an optimal model taking into account both linear and non-linear parts of variation of response variable, a strategy of hybrid modeling for combining nonparametric statistical model such as CART/MARS with linear regression analysis is proposed after the features are extracted from datasets. To valid the strategy, the datasets are applied, and useful results are obtained as follows. (1) Hybrid modeling can improve accuracy of the sole modeling of CART and/or regression model. (2) It can handle any case with surrogate variable instead of any predictor, which has a missing value. (3) Autocorrelation of residual time series and its residual mean squared error after hybrid can be improved by application of autoregressive model.

(4 results)