Anzeige
Erschienen in:
2016 | OriginalPaper | Buchkapitel
On the Relation Between kNN Accuracy and Dataset Compression Level
verfasst von : Marcin Blachnik
Erschienen in: Artificial Intelligence and Soft Computing
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
The paper discusses how instance selection can be used to asses the kNN performance. There exists a strong correlation between the compression level of the dataset obtained by instance selection methods and the prediction accuracy obtained the k-NN classifier trained on full training dataset.
Based on two standard algorithms of instance selection namely CNN and ENN, which belong to two different groups of methods, so called condensation and editing methods, we perform empirical analysis to verify this relation. The obtained results show that this relation is almost linear, so that the level of compression is linearly correlated with the accuracy. In other words by knowing the compression of instance selection methods we are able to estimate the accuracy of the final kNN prediction model.