| | |
Summary: Hybrid Voting Algorithms Using Selected Models for
Categorical Data
E. Graubins and D. Grossman
Department of Computer Science
Illinois Institute of Technology
Chicago Illinois 60616
{eug,grossman}@ir.iit.edu
ABSTRACT
Although voting methods are a viable way to improve
classification algorithm performance, these have usually been
applied to complete training datasets. We propose a new voting
methodology which is based on the success of each individual
classifier as it is applied to particular classes in a training dataset.
We test some specific variations on this theme and have found as
much as a 12.8% improvement in effectiveness over current
voting algorithms.
1. INTRODUCTION
Improving model effectiveness is a key goal of classification
algorithms. Voting algorithms, by combining results from
different classifiers, may outperform individual classifiers. Such
|
