Award Date

12-2010

Degree Type

Thesis

Degree Name

Master of Science in Electrical Engineering (MSEE)

Department

Electrical and Computer Engineering

First Committee Member

Ebrahim Saberinia, Chair

Second Committee Member

Shahram Latifi

Third Committee Member

Emma E. Regentova

Graduate Faculty Representative

Kazem Taghva

Number of Pages

122

Abstract

The advent of new technologies like DNA micro-arrays provides scientists the ability to gather important information such as the expression levels of almost all the genes within a cell. As the collected data is huge, it is always necessary to use analytical methods to extract important information which can be useful in biological and medical applications. One of such applications is presented in (Van‟t Veer LJ 2002), where the authors used the gene expression values obtained from micro-arrays of breast cancer cells to predict the outcome of the disease. The prediction is based on a supervised classification. While the idea of using gene expression values for breast cancer prognosis is very important, however the statistical methods used for designing the classifier were not chosen carefully. Therefore a thorough study of the problem can lead to an improved prognosis tool.

In this thesis, we concentrate on the classifier design for this problem. We examine and compare different feature selection methods such as Sequential forward selection (SFS), Sequential backward selection (SBS) and Bidirectional selection (BDS) and different classification algorithms such as Linear discriminant analysis (LDA), Quadratic discriminant analysis (QDA), Nearest Mean Classifier (NMC) and Support vector machines (SVM). We evaluated the error rate of these classifiers either with error estimation methods such as Resubstitution, Leave one out, Bootstrap or trying them on independent sample sets. Finally, we suggest a classifier for this problem and comparing its performance with the classifier proposed by (Van‟t Veer LJ 2002), show that our classifier performs much better in predicting the outcome of the disease.

Keywords

Genetic algorithms; Gene expression – Classification; Learning classifier systems

Disciplines

Biomedical | Electrical and Computer Engineering

Language

English


Included in

Biomedical Commons

Share

COinS