A comparison of soft fusion methods under different bagging scenarios

Authors:
Fuad Alkoot;Hussain Qasem
Affiliations:
-;-
Venue:
CIMMACS'07 Proceedings of the 6th WSEAS international conference on Computational intelligence, man-machine systems and cybernetics
Year:
2007

Citing 13
Cited 0

Original Contribution: Stacked generalization

Neural Networks
Decision Combination in Multiple Classifier Systems

IEEE Transactions on Pattern Analysis and Machine Intelligence
Combining the results of several neural network classifiers

Neural Networks
Bagging predictors

Machine Learning
Combination of Multiple Classifiers Using Local Accuracy Estimates

IEEE Transactions on Pattern Analysis and Machine Intelligence
On Combining Classifiers

IEEE Transactions on Pattern Analysis and Machine Intelligence
An Experimental Comparison of Three Methods for Constructing Ensembles of Decision Trees: Bagging, Boosting, and Randomization

Machine Learning
An Empirical Comparison of Voting Classification Algorithms: Bagging, Boosting, and Variants

Machine Learning
Sum Versus Vote Fusion in Multiple Classifier Systems

IEEE Transactions on Pattern Analysis and Machine Intelligence
Modified product fusion

Pattern Recognition Letters
A Brief Introduction to Boosting

IJCAI '99 Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence
Bagging in Computer Vision

CVPR '98 Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
Bagging, boosting, and C4.S

AAAI'96 Proceedings of the thirteenth national conference on Artificial intelligence - Volume 1

Quantified Score

Hi-index	0.00

Visualization

Abstract

We experiment with different fusion methods when bagging k-NN classifiers under various conditions. Experiments with four types of bagging are made at four training set sizes, using two metrics. The aim is to find the conditions for an optimum bagging performance. Additionally we aim to find the best rule under the specified conditions. We compare the performance of the different fusion strategies under each condition. Fusion methods used are Sum, Modified Product (MProduct) [2], Vote and Moderation [1]. Results show that the performance depends on the data used, number of nearest neighbors (k), number of fused classifiers and size of training set. Over all the three rules derived from Product show a close performance, while Vote shows an opposite performance. Among the three rules Moderation either follows Sum or MProduct. Results indicate MProduct outperforms Sum at many instances. At some of these instances Sum did not outperform the single classifier while MProduct did. Moderation is the second best, while Vote is inferior especially at even numbers of k. This is an inherited weakness of Vote, where at even number of close samples ties are randomly resolved. At k=1 all rules yield similar results. There are few instances where Moderation outperforms all. In general MProduct is the best choice.