| Grant number: | 11/14602-7 |
| Support type: | Scholarships in Brazil - Doctorate (Direct) |
| Effective date (Start): | November 01, 2011 |
| Effective date (End): | June 30, 2016 |
| Field of knowledge: | Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques |
| Principal Investigator: | André Carlos Ponce de Leon Ferreira de Carvalho |
| Grantee: | Luís Paulo Faina Garcia |
| Home Institution: | Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil |
| Associated scholarship(s): | 15/00741-6 - Metaheuristics for label noise identification in classification tasks, BE.EP.DD 13/20983-9 - Complexity measures for noise identification in classification problems, BE.EP.DD |
Abstract Data collected directly from storage systems often present high rate noise resulting from internal and external factors. When used in the induction of classifiers by machine learning techniques, these noisy data may reduce the predictive accuracy, increase the complexity of the hypothesis obtained and its induction time. This paper aims to investigate two research directions regarding this issue: noise class and predictive attributes noise. Methods like removal, reclassification and imputation of values are used to support the research, as well as machine learning and genetic algorithms based techniques. (AU) | |