Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
530347 | Pattern Recognition | 2014 | 10 Pages |
Abstract
Methods for tackling classification problems usually maximize prediction accuracy. However some applications require maximum predictive value instead. That is, the designer hopes to predict one of the classes with maximum precision, and is less concerned about the others. Some techniques exist for fine-tuning a model׳s predictive value, but there seems to be a shortage of methods to generate maximum-predictive-value classifiers. We propose a method using a nearest-prototype-style classifier optimized by a genetic algorithm. We test its performance using 13 publicly available data sets from the life sciences. The method generally gives more effective high-predictive-value models than standard classification methods optimized for predictive value.
Related Topics
Physical Sciences and Engineering
Computer Science
Computer Vision and Pattern Recognition
Authors
Eric Chalmers, Marcin Mizianty, Eric Parent, Yan Yuan, Edmond Lou,