Ensemble neural network based on attribute reduct and cluster analysis

Tao Zhou, Yan Ning Zhang, Hui Ling Lu, He Jin Yuan, Jun Li, Wei Wei

Research output: Contribution to journalArticlepeer-review

1 Scopus citations

Abstract

How to construct single neural network and improve classification performance in ensemble neural networks that possessing much difference is a key question. By studying the method of extract training samples possessing differences among feature space and sample, and a new sequence targets classification algorithm based on ensemble neural network about attributes reduct and cluster analysis was proposed. Because difference among samples is higher than before, classification performance of ensemble classifier is guaranteed effetely. Firstly, targets were extracted from sequence images, such as pedestrian, crowd, car from single frame image in training video. Profile features and geometrical features were acted as describing attributes about three kinds of targets; secondly, samples data possessing profile characteristics and geometrical characteristics were reducted by rough set theory and obtain three subspaces in feature space; thirdly, these samples were clustered using RPCL algorithm and obtain their distribution in sample space; fourthly, single neural network was constructed by this method proposed; finally, ensemble was constructed by relative majority voting method. Ensemble neural network based on boosting and bagging method were adopted to compare with this method. Experiment result illustrate that the method has a high classification precision comparing with tradition method, and it is a effective target classification approach.

Original languageEnglish
Pages (from-to)1365-1369
Number of pages5
JournalXitong Fangzhen Xuebao / Journal of System Simulation
Volume22
Issue number6
StatePublished - Jun 2010

Keywords

  • Attribute reduct
  • Cluster analysis
  • Ensemble NN
  • Sequence target

Fingerprint

Dive into the research topics of 'Ensemble neural network based on attribute reduct and cluster analysis'. Together they form a unique fingerprint.

Cite this