Flexible Classification Techniques with Biomedical Applications Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 19, 2019
  • Zhang, Chong
    • Affiliation: College of Arts and Sciences, Department of Statistics and Operations Research
  • Classification problems are prevalent in many scientific disciplines, especially in biomedical research. Recently, margin based classifiers have become increasingly popular, partly due to their ability in handling large scale problems with fast computational speed and desirable theoretical properties. Despite the success of margin based classifiers, many challenges remain. For example, in practical problems, it can be desirable to estimate the class conditional probability accurately. For high dimensional classification data, penalized margin based classifiers are commonly used. However, when estimating the class conditional probability, the shrinkage effect from the penalty term in the corresponding optimization is often ignored. This effect can lead to large bias in estimation of the class conditional probability. Another important issue on classification is the comparison between soft and hard classifiers for multicategory problems. Moreover, regular multicategory margin based classifiers can suffer from inefficiency by using too many classification functions. In this dissertation, we propose several new classification techniques to overcome the challenges mentioned above. Comprehensive numerical and theoretical studies are presented to demonstrate the usefulness of our new proposed methodologies.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Marron, James Stephen
  • Zeng, Donglin
  • Nobel, Andrew
  • Liu, Yufeng
  • Zhang, Kai
  • Doctor of Philosophy
Degree granting institution
  • University of North Carolina at Chapel Hill Graduate School
Graduation year
  • 2014
Place of publication
  • Chapel Hill, NC
  • This item is restricted from public view for 1 year after publication.

This work has no parents.