Improvement of Prediction Accuracy Using Discretization and Voting Classifier

There are many examples of classification algorithms developed so far for data analysis, pattern recognition, scene analysis and learning from graphical models. Being motivated by the works of a number of researchers, here the author have tried to improve the prediction accuracy by first discretizin...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:18th International Conference on Pattern Recognition (ICPR'06) Ročník 2; s. 695 - 698
Hlavní autor: Ekbal, A.
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 2006
Témata:
Accuracy
Ada-Boost Algorithm
Bagging
Bayesian classifier
Bayesian methods
Boosting
Classification
Classification algorithms
Data analysis
Decision trees
Naïve
Neural networks
Pattern recognition
SPID3 Algorithm
Voting
Voting classifier
ISBN:0769525210, 9780769525211
ISSN:1051-4651
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:There are many examples of classification algorithms developed so far for data analysis, pattern recognition, scene analysis and learning from graphical models. Being motivated by the works of a number of researchers, here the author have tried to improve the prediction accuracy by first discretizing the real world dataset and then applying a voting classifier on the discretized dataset. In this work, continuous dataset from the raw real world dataset having missing attribute values have been generated and discretized the dataset using SPID 3 algorithm. Then naive-Bayesian classifier has been implemented to apply it on the continuous and discretized dataset. Finally, an ensemble learner (Ada-boost algorithm) has been developed where the naive Bayesian classifier has been used as the base learner of the ensemble. The extensive empirical results over the twenty real world datasets show that the prediction accuracy can be increased by the joint performance of discretization and voting classifier
ISBN:0769525210
9780769525211
ISSN:1051-4651
DOI:10.1109/ICPR.2006.698