Small-Sample Classification and Error Estimation

May 4th, 2014 | By | Category: Other, Systems Engineering

Small-Sample Classification and Error EstimationResearchers: Ulisses Braga-Neto

The research goals of this proposal concern the solution of significant computational and statistical problems in classification error estimation, with the purpose of improving the assessment of predictions made in classification and inference of genomic and proteomic signals based on small samples in high-dimensional spaces. This research critically impacts the discovery of reliable molecular biomarkers for disease diagnosis and prognosis in Genomics and Proteomics applications that rely on pattern recognition approaches. Some of the highlights of this research include the development of a small-RMS, small-sample, kernel-based error estimator (“boltered error estimation”), the detailed study of the variance of cross-validation methods in small-sample settings, and the analytical study of higher moments and sampling distributions of error estimators for Linear Discriminant Analysis.

Comments are closed.