A Case Study of SVM Extension Techniques on classification of Imbalanced Data

K. K. Lee; C. J. Harris; S. R. Gunn; P. A. S. Reed

A Case Study of SVM Extension Techniques on classification of Imbalanced Data

K. K. Lee, C. J. Harris, S. R. Gunn, P. A. S. Reed

School of Computing Sciences

Research output: Contribution to conference › Paper

Abstract

In many classification problems the data is imbalanced, that is the class priors are different. Here, we consider the classification problem of fatigue crack initiation in automotive camshafts, where this imbalance is significant. The extension techniques of Support Vector Machine (SVM) - the Control Sensitivity (CSSVM) and Adaptive Margin (AMSVM) - which offer different ways of dealing with imbalanced data was investigated. Geometric mean was used to evaluate the performance of the model. The CSSVM has outperformed the AMSVM. The use of different kernels did not produce significant changes in the results. The ratio between the misclassification cost and the training size for each class is very similar, indicating a strong relationship between them.

Original language	English
Pages	309-314
Number of pages	6
Publication status	Published - Feb 2001
Event	Congress on Neural Networks and Applications, Fuzzy Sets and Fuzzy Systems and Evolutionary Computing - Puerto De La Cruz, Tenerife, Canary Islands, Spain Duration: 11 Feb 2001 → 15 Feb 2001

Conference

Conference	Congress on Neural Networks and Applications, Fuzzy Sets and Fuzzy Systems and Evolutionary Computing
Country/Territory	Spain
City	Puerto De La Cruz, Tenerife, Canary Islands
Period	11/02/01 → 15/02/01

Cite this

@conference{6f4a4233ee5c41099b25640351732c2c,

title = "A Case Study of SVM Extension Techniques on classification of Imbalanced Data",

abstract = "In many classification problems the data is imbalanced, that is the class priors are different. Here, we consider the classification problem of fatigue crack initiation in automotive camshafts, where this imbalance is significant. The extension techniques of Support Vector Machine (SVM) - the Control Sensitivity (CSSVM) and Adaptive Margin (AMSVM) - which offer different ways of dealing with imbalanced data was investigated. Geometric mean was used to evaluate the performance of the model. The CSSVM has outperformed the AMSVM. The use of different kernels did not produce significant changes in the results. The ratio between the misclassification cost and the training size for each class is very similar, indicating a strong relationship between them.",

author = "Lee, {K. K.} and Harris, {C. J.} and Gunn, {S. R.} and Reed, {P. A. S.}",

year = "2001",

month = feb,

language = "English",

pages = "309--314",

note = "Congress on Neural Networks and Applications, Fuzzy Sets and Fuzzy Systems and Evolutionary Computing ; Conference date: 11-02-2001 Through 15-02-2001",

}

TY - CONF

T1 - A Case Study of SVM Extension Techniques on classification of Imbalanced Data

AU - Lee, K. K.

AU - Harris, C. J.

AU - Gunn, S. R.

AU - Reed, P. A. S.

PY - 2001/2

Y1 - 2001/2

N2 - In many classification problems the data is imbalanced, that is the class priors are different. Here, we consider the classification problem of fatigue crack initiation in automotive camshafts, where this imbalance is significant. The extension techniques of Support Vector Machine (SVM) - the Control Sensitivity (CSSVM) and Adaptive Margin (AMSVM) - which offer different ways of dealing with imbalanced data was investigated. Geometric mean was used to evaluate the performance of the model. The CSSVM has outperformed the AMSVM. The use of different kernels did not produce significant changes in the results. The ratio between the misclassification cost and the training size for each class is very similar, indicating a strong relationship between them.

AB - In many classification problems the data is imbalanced, that is the class priors are different. Here, we consider the classification problem of fatigue crack initiation in automotive camshafts, where this imbalance is significant. The extension techniques of Support Vector Machine (SVM) - the Control Sensitivity (CSSVM) and Adaptive Margin (AMSVM) - which offer different ways of dealing with imbalanced data was investigated. Geometric mean was used to evaluate the performance of the model. The CSSVM has outperformed the AMSVM. The use of different kernels did not produce significant changes in the results. The ratio between the misclassification cost and the training size for each class is very similar, indicating a strong relationship between them.

M3 - Paper

SP - 309

EP - 314

T2 - Congress on Neural Networks and Applications, Fuzzy Sets and Fuzzy Systems and Evolutionary Computing

Y2 - 11 February 2001 through 15 February 2001

ER -