Supervised classifiers based on emerging patterns for class imbalance problems
In the last years, emerging pattern-based classifiers have become an important family of supervised classifiers. However, in those problems where the objects are not equally distributed into the classes (class imbalance problems), emerging pattern mining algorithms, not designed for this kind of pro...
| Autor: | |
|---|---|
| Formato: | tesis doctoral |
| Estado: | Versión aceptada para publicación |
| Fecha de publicación: | 2017 |
| País: | México |
| Recursos: | Instituto Nacional de Astrofísica, Óptica y Electrónica |
| Repositorio: | Repositorio Institucional del INAOE |
| Idioma: | inglés |
| OAI Identifier: | oai:inaoe.repositorioinstitucional.mx:1009/826 |
| Acesso em linha: | http://inaoe.repositorioinstitucional.mx/jspui/handle/1009/826 |
| Access Level: | acceso abierto |
| Palavra-chave: | info:eu-repo/classification/Reconocimiento de patrones/Pattern recognition info:eu-repo/classification/Problemas de desequilibrio/Imbalance problems info:eu-repo/classification/Clasificadores supervisados/Supervised classifiers info:eu-repo/classification/cti/1 info:eu-repo/classification/cti/12 info:eu-repo/classification/cti/1203 info:eu-repo/classification/cti/120323 |
| Resumo: | In the last years, emerging pattern-based classifiers have become an important family of supervised classifiers. However, in those problems where the objects are not equally distributed into the classes (class imbalance problems), emerging pattern mining algorithms, not designed for this kind of problems, extract several emerging patterns with high support for the majority class and only a few (or none) emerging patterns with low support for the minority class. As a consequence, emerging pattern-based classifiers tend to bias their classification results toward the majority class; obtaining poor classification results for the minority class. Hence, in this PhD research, we first present a study about the effect of class imbalance on quality measures for patterns; from this study, we select the best measure for ranking emerging patterns in class imbalance problems. Additionally, we propose three new algorithms for extracting emerging patterns from imbalanced databases. Our emerging pattern mining algorithms extract a collection of emerging patterns which allows attaining higher accuracies for supervised classification in class imbalance problems than those emerging patterns extracted by other emerging pattern miners developed for this kind of problems. Finally, we propose a new emerging pattern-based classifier specifically designed for class imbalance problems, which obtains significantly better classification results than other classifiers for class imbalance problems reported in the literature. |
|---|