An Ensemble of Classifiers using Dynamic Method on Ambiguous Data
Author(s):
Dnyaneshwar Kudande
Keywords:
Datamining, data reduction,instance selection, Classification, weightedinstance selection,ReducedNearest Neighbor.
Abstract
Theaim ofproposed work istoanalyzethe Instance Selection Algorithmfirst. ThereareWeighted Instance Selection algorithms areavailablesuch aswDROP3 (weighted DecrementalReduction Optimization Procedure 3), wRNN(weighted Reduced NearestNeighbor),which reduces theSamplesetapplied.Then themulticlassInstance Selection isuseful techniqueforreducingspace and timecomplexity. This removes irrelevant, noisy,superfluous instances from TrainingSet.Then themulticlassproblem issolvedby consideringnumberoftwo classproblem thusdesigning multiple two class classifiers and its combined output produces the resultfor it.The Boostingisuse for providing weightforeach instanceoftrainingset.TheDesigningof ensembleofclassifiersistocombineallclassifiersandlearn byreduced training set. There aredifferenttechniquesare available for designing an ensemble such as Bagging (Bootstrap Aggregating), Boosting(ADABOOST)and Error CorrectingOutputCode (ECOC) etc.Theoutputof ensemble isbetter than theindividualclassifiers.Theapproach istested with fewbenchmark datasets.Itis found thatClassification accuracyinthecaseofwDROP3algorithm liesbetween70% to87%,butincaseofwRNNalgorithmliesbetween61%to89%andtheGeneralizationaccuracyinthecaseofwDROP3 algorithmliesbetween79%to96%,butinwRNNalgorithmit lies between 75% to94%. Another observation, whenincreasesnumberofClassifiersperEnsemblethenaccuracy improves by0.5to1.5%.
Article Details
Unique Paper ID: 142548

Publication Volume & Issue: Volume 2, Issue 3

Page(s): 88 - 95
Article Preview & Download


Share This Article

Join our RMS



Conference Alert

NCSEM 2024

National Conference on Sustainable Engineering and Management - 2024

Last Date: 15th March 2024

Latest Publication

Call For Paper

Volume 10 Issue 10

Last Date for paper submitting for March Issue is 25 June 2024

About Us

IJIRT.org enables door in research by providing high quality research articles in open access market.

Send us any query related to your research on editor@ijirt.org

Social Media

Google Verified Reviews

Contact Details

Telephone:6351679790
Email: editor@ijirt.org
Website: ijirt.org

Policies

Important Links

Browse Rsearch papers