Presentation 1999/3/19
A New BP Classification Method Using Multi Variate Analysis
Nobukatsu Kitajima,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, a new Back Propagation (BP) classification method is proposed for attributive data like customer data to use at credit scoring or marketing such as age, income, occupation and so on. The method employs the number of input predictor variables which effects on classification results seriously, as the least number of middle unit of multi layer perceptron (MLP). In the method, data translation into the form which is easier to learn for MLP is also executed. From numerical simulations using credit scoring data, the classifier is able to prevent over-fitting for delinquent data which is supposed as having complicated distribution. The method showed 20% higher classification performance in delinquent data recognition of completely unknown data without serious loss for whole classification performance than quantification theory type II classifier which is used conventionally.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) BP / MLP / multivariate analysis / linear discriminant analysis / classification / data mining
Paper # NC98-153
Date of Issue

Conference Information
Committee NC
Conference Date 1999/3/19(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 Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) A New BP Classification Method Using Multi Variate Analysis
Sub Title (in English)
Keyword(1) BP
Keyword(2) MLP
Keyword(3) multivariate analysis
Keyword(4) linear discriminant analysis
Keyword(5) classification
Keyword(6) data mining
1st Author's Name Nobukatsu Kitajima
1st Author's Affiliation Human Media Research Laboratories, NEC Corporation()
Date 1999/3/19
Paper # NC98-153
Volume (vol) vol.98
Number (no) 674
Page pp.pp.-
#Pages 8
Date of Issue