Metaheuristics for label noise identification in classification tasks

• Grant number: 15/00741-6
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate (Direct)
• Start date: May 01, 2015
• End date: June 11, 2015
• 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
• Supervisor: Stan Matwin
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Dalhousie University, Halifax, Canada
• Associated to the scholarship: 11/14602-7 - Noise detection and elimination for classification problems, BP.DD

Abstract

The successful application of machine learning techniques to data sets from many areas requires the dedication of a considerable amount of time to identify and treat noisy data. Noisy data is a frequent problem regarding information collection, transmission and storage. The presence of noise in data can affect the complexity of classification tasks, making the discrimination of objects from different classes more difficult, and requiring more complex decision boundaries for classification model. In this project, we will study the effects of noise in the complexity of classification data sets by using geometric and statistical measures extracted from a data set. Those indices more sensitive to the presence of noise can then be used in the proposal of metaheuristics capable of searching for label noise in classification problems. (AU)