Xiaoke Fei

Type

Text

Type

Dissertation

Advisor

Ahn, Hongshik. | Zhu, Wei | Wu, Song | Park, Memming.

Date

2017-08-01

Keywords

Statistics | CERP | Classification | Ensemble | Random Forest | WAVE

Department

Department of Applied Mathematics and Statistics.

Language

en_US

Source

This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree

Identifier

http://hdl.handle.net/11401/78126

Publisher

The Graduate School, Stony Brook University: Stony Brook, NY.

Format

application/pdf

Abstract

This dissertation developed two weighted voting classification ensemble methods: Weight-Adjusted CERP (WACERP) and Weight-Adjusted Random Forest (WARF). WACERP, built by applying the WAVE voting algorithm to CERP, is an ensemble method designed specially for high-dimensional data sets. Our study used two high-dimensional data sets to investigate the performance of WACERP. The result showed that WACERP consistently outperforms CERP in terms of accuracy, as well as maintaining balance between sensitivity and specificity. WACERP also performs consistently well compared to other popular classification methods. WARF is built by applying WAVE to Random Forest (RF). To evaluate the performance of WARF, we applied WARF, RF and some other widely used classification models to 23 data sets from various areas. Our study showed that WARF performs consistently better than RF and the other classification methods. WARF achieves its best performance at lower ensemble size than RF in general. | 108 pages

Download

Share

COinS