A self-generating prototype method based on information entropy used for condensing data in classification tasks

Manastarla A.; Silva L.A.

A self-generating prototype method based on information entropy used for condensing data in classification tasks

item.page.type

Artigo de evento

Date

2019

item.page.ispartof

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

item.page.citationsscopus

0

Authors

Manastarla A.
Silva L.A.

Abstract

© 2019, Springer Nature Switzerland AG.This paper presents a new self-generating prototype method based on information entropy to reduce the size of training datasets. The method accelerates the classifier training time without significantly decreasing the quality in the data classification task. The effectiveness of the proposed method is compared to the K-nearest neighbour classifier (kNN) and the genetic algorithm prototype selection (GA). kNN is a benchmark method used for data classification tasks, while GA is a prototype selection method that provides competitive optimisation of accuracy and the data reduction ratio. Considering thirty different public datasets, the results of the comparisons demonstrate that the proposed method outperforms kNN when using the original training set as well as the reduced training set obtained via GA prototype selection.

item.page.scopussubject

Classification tasks , Classifier training , Data classification , Information entropy , K-nearest neighbours , Prototype selection , Reduced training sets , Training data sets