A genetic algorithm for hierarchical multi-label classification problems

• Grant number: 11/22321-8
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: December 01, 2012
• End date: May 31, 2013
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: André Carlos Ponce de Leon Ferreira de Carvalho
• Grantee: Ricardo Cerri
• Supervisor: Alex A. Freitas
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: University of Kent, England
• Associated to the scholarship: 09/17401-2 - Investigation of Bioinspired Approaches for Hierarchical Multilabel Classification Problems, BP.DR

Abstract

In conventional classification problems, each example of a dataset can be associated with only one of two or more classes. However, there are more complex classification problems, where the classes involved are hierarchically structured, possibly including subclasses and superclasses. These problems are known as hierarchical classification problems, as examples can be associated to classes belonging to a path of a class hierarchy. Such a hierarchy can be structured as a tree or as a directed acyclic graph. Among the hierarchical problems, there are those in which examples can be simultaneously assigned to classes belonging to two or more paths of a class hierarchy, i.e., examples can be classified into several classes located in the same hierarchical level. These problems are called hierarchical multi-label classification problems and can have hundreds or even thousands of classes. In addition, these problems have characteristics that make it difficult to use conventional classification methods, such as high complexity, diversity of solutions and difficult modeling. These characteristics can be best dealt using bioinspired approaches such as genetic algorithms, because these are composed of more efficient probabilistic methods, and globally seek for one or more solutions to search and optimization problems. Thus, during the period of research abroad, a genetic algorithm for generating hierarchical and multi-label classification rules will be proposed. (AU)