Author:
Dymitr Ruta
Affiliation:
British Telecom, Research & Venturing, United Kingdom
Keyword(s):
Classification, feature selection, classifier fusion, genetic algorithm.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Artificial Intelligence and Decision Support Systems
;
Biomedical Engineering
;
Business Analytics
;
Data Engineering
;
Data Mining
;
Databases and Information Systems Integration
;
Datamining
;
Enterprise Information Systems
;
Evolutionary Programming
;
Health Information Systems
;
Sensor Networks
;
Signal Processing
;
Soft Computing
Abstract:
Recent research efforts dedicated to classifier fusion have made it clear that combining performance strongly depends on careful selection of classifiers. Classifier performance depends, in turn, on careful selection of features, which on top of that could be applied to different subsets of the data. On the other hand, there is already a number of classifier fusion techniques available and the choice of the most suitable method relates back to the selection in the classifier, feature and data spaces. Despite this apparent selection multidimensionality, typical classification systems either ignore the selection altogether or perform selection along only single dimension, usually choosing the optimal subset of classifiers. The presented multidimensional selection sketches the general framework for the optimised selection carried out simultaneously on many dimensions of the classification model. The selection process is controlled by the specifically designed genetic algorithm, guided d
irectly by the final recognition rate of the composite classifier. The prototype of the 3-dimensional fusion-classifier-feature selection model is developed and tested on some typical benchmark datasets.
(More)