Complexity measures for noise identification in classification problems

• Grant number: 13/20983-9
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate (Direct)
• Start date: March 20, 2014
• End date: September 19, 2014
• 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: Francisco Herrera
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Universidad de Granada (UGR), Spain
• Associated to the scholarship: 11/14602-7 - Noise detection and elimination for classification problems, BP.DD

Abstract

Many areas of knowledge have spent considerable amounts of time to comprehend and to treat noisy data, one of the most common problems regarding information collection, transmission and storage. In computer studies, it occurs mainly in real data collected from storage systems, appearing in high rates. These noisy data, when used in the classifiers induction by Machine Learning techniques, increase the complexity of the obtained hypothesis, as well as its induction time while worsening their accuracy. In classification problems, noises are treated from two different perspectives: noises in predictive attributes and noises in the target attribute. Treating them may improve the data quality. Considering this background, the central aim of this project is the study of complexity measures capable of recognizing noisy patterns which may enable the creation of search tools for the identification of noisy data. (AU)