Abdullah, , and Ku-Mahamud, Ku Ruhana (2012) Prediction accuracy measurements for ensemble classifier. In: Knowledge Management International Conference (KMICe) 2012, 4 – 6 July 2012, Johor Bahru, Malaysia.

Preview

PDF Download (260kB) | Preview

Abstract

Multiple classifier combination (or ensemble method) has been shown to be very helpful in improving the performance of classification over single classifier approach. The diversity among base classifiers (or ensemble members) is important when constructing a classifier ensemble.Although there have been several measures of diversity, but there is no reliable measure that can predict the ensemble accuracy. The base classifiers accuracy will increase when the diversity decreases and this is known as the accuracy-diversity dilemma.This paper presents a new method to measure diversity in classifier ensembles.Furthermore another parameter which based on this diversity measure is defined.It is hope that the new parameter will be able to predict the ensemble accuracy.Based on experimental results on classification of 84 samples of fruit images using nearest mean classifier ensembles, it has been shown that there is a positive linear relationship between the new parameter and the ensemble accuracy.This parameter is expected to assist in constructing diverse and accurate ensemble.

Item Type:	Conference or Workshop Item (Paper)
Additional Information:	Organized by Universiti Utara Malaysia
Uncontrolled Keywords:	Classification, multiple classifier combination, diversity measure.
Subjects:	Q Science > QA Mathematics > QA76 Computer software
Divisions:	College of Arts and Sciences
Depositing User:	Prof. Dr. Ku Ruhana Ku Mahamud
Date Deposited:	10 Jan 2013 01:57
Last Modified:	25 May 2015 01:27
URI:	https://repo.uum.edu.my/id/eprint/6972

Actions (login required)

View Item