Distribution-Based Classification with Application to Gene Expression Data Analysis

• Activity Number: 196
• Type: Contributed
• Date/Time: Monday, July 30, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xuelian Wei*+ and Ker-Chau Li
• Companies: University of California, Los Angeles and University of California, Los Angeles
• Keywords: Classification ; Prediction ; Machine learning ; Cancer classfication ; Microarray ; Gene expression data

Abstract

Distribution-based classification (DBC) is a new technique for class prediction. Unlike univariate classification methods, such as gene voting, which treat each variable separately, DBC considers the joint effects of all variables simultaneously. Unlike K-nearest neighbor method (KNN) or Support vector machine (SVM), which utilize only a small proportion of training samples (the K nearest neighbors or the support vectors), DBC exploits the joint prediction strengths of all training samples. We present a geometric configuration to illustrate when our method have a clear advantage over other methods. When applying to 22 gene expression datasets for cancer classification, DBC performs as efficiently as the SVM and naive Bayesian. All outperform other learning methods (KNN, linear discriminant analysis, and decision tree).