Summary
One of the approaches adopted to generate multiclass classifiers from binary predictors is to decompose the multiclass problem into multiple binary subproblems. Among the existing decomposition approaches, one may cite the use of Directed Acyclic Graphs (DAG) to combine pairwise classifiers. This work presents a study on the influence of the DAG structure in the performance obtained in multiclass problems when Support Vector Machines are used in the induction of the binary predictors.
See the full content of this document
Extract
Design of Directed Acyclic Graph Multiclass Structures
1. Introduction
Several Machine Learning (ML) [17] are originally designed to solve binary classification problems. Among then, one may cite the Support Vector Machines (SVMs) [3]. Two approaches may be followed to solve multiclass classification problems with binary predictors. The first approach involves adapting the learning technique training algorithm in order to solve the multiclass problem directly. Hsu and Lin [12] point out that for SVMs in particular the training algorithms obtained have a high computational cost. The second approach consists of decomposing the multiclass problem into multiple binary subproblems and combine their solution in order to solve the original problem. This second approach is employed more often.One of the existing decomposition strategies consists of disposing binary classifiers for pairs of classes into a Directed Acyclic Graph (DAG) [21, 13]. Some studies [13, 26, 7] show that the order of the classifiers in this graph may influence the performance achieved in the multiclass problem solution. This work investigates the use of Genetic Algorithms (GAs) [16] in the design of the DAG structures for the rnulticlass problems when the SVMs are used to generate the binary classifiers.This paper is structured as follows: section 2 briefly describes the main aspects of the SVMs. section 3 presents how DAGs can be used in order to obtain multiclass predictions with binary classifiers. section 4 covers, the GAs and discusses how they ...See the full content of this document
Sponsored links
