Standoff-Balancing: A Novel Class Imbalance Treatment Method Inspired by Military Strategy
M. Siers, and M. Islam. AI 2015: Advances in Artificial Intelligence, volume 9457 of Lecture Notes in Computer Science, Springer International Publishing, (2015)
Abstract
A class imbalanced dataset contains a disproportionate number of a certain class’ records compared to other classes. Classifiers which are built from class imbalanced datasets are biased and thus under-perform for the minority class. Treatment methods such as sampling and cost-sensitivity can be used to negate the bias induced by class imbalance. In this study, we present an analogy between class imbalance and war. By creating this analogy, we make it possible for military strategies to be applied to class imbalanced datasets. We propose a novel class imbalance treatment method Standoff-Balancing which uses a well-known mathematical law from military strategy literature. We compare the proposed technique with four existing techniques on five real world data sets. Our experiments show that the proposed technique may provide a higher AUC to existing techniques.
%0 Book Section
%1 siers2015standoff
%A Siers, Michael J
%A Islam, Md Zahidul
%B AI 2015: Advances in Artificial Intelligence
%D 2015
%E Pfahringer, Bernhard
%E Renz, Jochen
%I Springer International Publishing
%K class-imbalance classification cost-sensitive myown
%P 517--525
%T Standoff-Balancing: A Novel Class Imbalance Treatment Method Inspired by Military Strategy
%U http://dblp.uni-trier.de/db/conf/ausai/ausai2015.html#SiersI15
%V 9457
%X A class imbalanced dataset contains a disproportionate number of a certain class’ records compared to other classes. Classifiers which are built from class imbalanced datasets are biased and thus under-perform for the minority class. Treatment methods such as sampling and cost-sensitivity can be used to negate the bias induced by class imbalance. In this study, we present an analogy between class imbalance and war. By creating this analogy, we make it possible for military strategies to be applied to class imbalanced datasets. We propose a novel class imbalance treatment method Standoff-Balancing which uses a well-known mathematical law from military strategy literature. We compare the proposed technique with four existing techniques on five real world data sets. Our experiments show that the proposed technique may provide a higher AUC to existing techniques.
@incollection{siers2015standoff,
abstract = {A class imbalanced dataset contains a disproportionate number of a certain class’ records compared to other classes. Classifiers which are built from class imbalanced datasets are biased and thus under-perform for the minority class. Treatment methods such as sampling and cost-sensitivity can be used to negate the bias induced by class imbalance. In this study, we present an analogy between class imbalance and war. By creating this analogy, we make it possible for military strategies to be applied to class imbalanced datasets. We propose a novel class imbalance treatment method Standoff-Balancing which uses a well-known mathematical law from military strategy literature. We compare the proposed technique with four existing techniques on five real world data sets. Our experiments show that the proposed technique may provide a higher AUC to existing techniques.},
added-at = {2016-06-08T12:59:29.000+0200},
author = {Siers, Michael J and Islam, Md Zahidul},
biburl = {https://www.bibsonomy.org/bibtex/25e01dda242c9a18ca42bcb77fc7aea4d/mikesiers},
booktitle = {AI 2015: Advances in Artificial Intelligence},
editor = {Pfahringer, Bernhard and Renz, Jochen},
interhash = {876901db979e3cffb5a027c110e41bc5},
intrahash = {5e01dda242c9a18ca42bcb77fc7aea4d},
keywords = {class-imbalance classification cost-sensitive myown},
pages = {517--525},
publisher = {Springer International Publishing},
series = {Lecture Notes in Computer Science},
timestamp = {2016-06-09T09:16:47.000+0200},
title = {Standoff-Balancing: A Novel Class Imbalance Treatment Method Inspired by Military Strategy},
type = {Publication},
url = {http://dblp.uni-trier.de/db/conf/ausai/ausai2015.html#SiersI15},
volume = 9457,
year = 2015
}