A Lexicographic Genetic Algorithm for Hierarchical Classification Rule Induction

Author(s)

Pereira, Gean Trindade ; Gabriel, Paulo. H. R. ; Cerri, Ricardo ; LopezIbanez, M

• Document type: Journal article
• Source: PROCEEDINGS OF THE 2019 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE (GECCO'19); v. N/A, p. 9-pg., 2019-01-01.

Abstract

Hierarchical Classification (HC) consists of assigning an instance to multiple classes simultaneously in a hierarchical structure containing dozens or even hundreds of classes. A field that greatly benefits from HC is Bioinformatics, in which interpretable methods that make predictions automatically are still scarce. In this context, a topic that has gained attention is the classification of Transposable Elements (TEs), which are DNA fragments capable of moving inside the genome of their hosts, affecting the genes' functionalities in many species. Thus, in this paper, we propose a method called Hierarchical Classification with a Lexicographic Genetic Algorithm (HC-LGA), which evolves rules towards the HC of TEs. Our proposed method follows a Multi-Objective Lexicographic approach in order to better deal with the still relevant problem of accuracy-interpretability trade-off. Besides, to the best of our knowledge, this is the first work to combine HC with such an approach. Experiments with two popular TEs datasets showed that HC-LGA achieved competitive results compared with most of the state-of-the-art HC methods in the literature, having the advantage of generating an interpretable model. Furthermore, HC-LGA obtained a comparable performance against its simpler optimization version, however generating a more interpretable list of rules. (AU)

• FAPESP's process: 15/14300-1 - Hierarchical classification of transposable elements using machine learning
• Grantee: Ricardo Cerri
• Support Opportunities: Regular Research Grants
• FAPESP's process: 16/50457-5 - Multi-objective evolutionary methods for hierarchical and multi-label classification
• Grantee: Ricardo Cerri
• Support Opportunities: Regular Research Grants