TY - GEN

T1 - System evaluation of construction methods for multi-class problems using binary classifiers

AU - Hirasawa, Shigeichi

AU - Kumoi, Gendo

AU - Kobayashi, Manabu

AU - Goto, Masayuki

AU - Inazumi, Hiroshige

N1 - Funding Information:
Acknowledgment. One of the authors S. H. would like to thank Professor Shin’ ichi Oishi of Waseda University for giving a chance to study this work. The research leading to this paper was partially supported by MEXT Kakenhi under Grant-in Aids for Scientific Research (B) No. 26282090 and (C) No. 16K00491.

PY - 2018

Y1 - 2018

N2 - Construction methods for multi-valued classification (multi-class) systems using binary classifiers are discussed and evaluated by a trade-off model for system evaluation based on rate-distortion theory. Suppose the multi-class systems consisted of M(≥3) categories and N(≥M-1) binary classifiers, then they can be represented by a matrix W, where the matrix W is given by a table of M code words with length N, called a code word table. For a document classification task, the relationship between the probability of classification error Pe and the number of binary classifiers N for given M is investigated, and we show that our constructed systems satisfy desirable properties such as “Flexible”, and “Elastic”. In particular, modified Reed Muller codes perform well: they are shown to be “Effective elastic”. As a second application we consider a hand-written character recognition task, and we show that the desirable properties are also satisfied.

AB - Construction methods for multi-valued classification (multi-class) systems using binary classifiers are discussed and evaluated by a trade-off model for system evaluation based on rate-distortion theory. Suppose the multi-class systems consisted of M(≥3) categories and N(≥M-1) binary classifiers, then they can be represented by a matrix W, where the matrix W is given by a table of M code words with length N, called a code word table. For a document classification task, the relationship between the probability of classification error Pe and the number of binary classifiers N for given M is investigated, and we show that our constructed systems satisfy desirable properties such as “Flexible”, and “Elastic”. In particular, modified Reed Muller codes perform well: they are shown to be “Effective elastic”. As a second application we consider a hand-written character recognition task, and we show that the desirable properties are also satisfied.

KW - Binary classifier

KW - ECOC

KW - Error correcting codes

KW - Exhaustive code

KW - Multi-valued classification

KW - Trade-off model

UR - http://www.scopus.com/inward/record.url?scp=85045309941&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045309941&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-77712-2_86

DO - 10.1007/978-3-319-77712-2_86

M3 - Conference contribution

AN - SCOPUS:85045309941

SN - 9783319777115

T3 - Advances in Intelligent Systems and Computing

SP - 909

EP - 919

BT - Trends and Advances in Information Systems and Technologies

A2 - Reis, Luis Paulo

A2 - Rocha, Alvaro

A2 - Costanzo, Sandra

A2 - Adeli, Hojjat

PB - Springer Verlag

T2 - 6th World Conference on Information Systems and Technologies, WorldCIST 2018

Y2 - 27 March 2018 through 29 March 2018

ER -